JPH11275582A - Image encoding device, image decoding device, and methods thereof - Google Patents

Abstract

(57) Abstract: A high-quality encoded and decoded image with low block distortion is obtained for a still image or a moving image even at a high compression ratio. SOLUTION: The orthogonal transform coefficients are divided into a plurality of zones ZONE1 to ZONE4 having different statistical properties, and the coefficients are individually encoded for each of the zones ZONE1 to ZONE4, thereby reducing the amount of encoding bits. Can be.
Also, a plurality of scalar quantization functions D ₀ , D ₁ , D ₂ , D
By using the trellis quantization means 9 having ₃
The quantization error can be reduced. Further, the image is divided into a plurality of band components, and each band image is divided into a zone ZONE1.
By performing scanning while dividing into ZONE4 and performing trellis quantization, compression encoding and decoding can be more easily and efficiently performed.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[Table of Contents] The present invention will be described in the following order.

BACKGROUND OF THE INVENTION Prior Art (FIG. 25) Problems to be Solved by the Invention Means for Solving the Problems Embodiments of the Invention (1) First Embodiment (FIGS. 1 to 3) (2) Second embodiment (FIGS. 4 to 8) (3) Third embodiment (FIGS. 9 and 10) (4) Fourth embodiment (FIGS. 11 and 12) (5) ) Fifth Embodiment (FIGS. 13 to 16) (6) Sixth Embodiment (FIGS. 17 to 19) (7) Seventh Embodiment (FIG. 20) (8) Eighth Embodiment Embodiment (FIG. 21) (9) Ninth Embodiment (FIGS. 22 and 23) (10) Other Embodiment (FIG. 24) Effects of the Invention

[0003]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image encoding apparatus, an image decoding apparatus, and a method thereof, for example, an image encoding apparatus and an image decoding apparatus used for transmitting a high-definition image via a communication satellite, and the like. It is suitable for application to the method of (1).

[0004]

2. Description of the Related Art As a conventional representative image compression method, I
JPEG (Joint Photographi) standardized by SO
c Experts Group) method. This is DCT (Discrete
If a relatively large number of bits are assigned using CosineTransform), encoded and decoded images with little deterioration can be obtained.

However, when the number of coding bits is reduced in the DCT processing, block distortion peculiar to DCT is remarkably generated, and image deterioration becomes noticeable. This DCT
As a method of eliminating the specific block distortion, for example, as disclosed in Japanese Patent Application Laid-Open No. 7-50835, an image is divided into sub-bands, and a moving image is obtained by performing orthogonal transformation on the divided low-frequency components. Encode. According to the method of subband division in this way, by utilizing the fact that the energy of the image signal is concentrated in the low band, more coding bits are allocated to the low band signal, and visually high-quality code can be obtained. It is considered that a coded image can be obtained with a smaller bit length.

That is, FIG. 25 shows an image coding apparatus for performing orthogonal transformation by dividing an image into sub-bands. A hybrid transformer 213 inputs an original image signal S 300 to a sub-band divider 200. The sub-band divider 200 divides the original image signal S300 into a low-band image and a plurality of high-band images by sub-band division, and sends this to the orthogonal transformer 201 and the block divider 202 as a sub-band division output signal S301.

The orthogonal transformer 201 divides the low-frequency image of the sub-band divided output signal S301 into blocks of a first size and orthogonally transforms the resultant, thereby obtaining an orthogonally transformed output signal S302.
Get. The block divider 202 divides a plurality of high-frequency images obtained by the sub-band divider 200 into blocks of a first size. And the block divider 20
2 combines the orthogonally transformed block of the first size with the block of the first size of the high-frequency image to form a block of the second size, thereby performing the hybrid transform. Do.

In the hybrid inverse transformer 214, the frame memory 210 rearranges the hybrid transform coefficient signal S303 to create a low-frequency image and a plurality of high-frequency images, and the inverse orthogonal transformer 211 outputs the low-frequency image Is divided into blocks of a first size and inverse orthogonally transformed. The subband synthesizer 212 performs subband synthesis on the entire image. Thus, in the hybrid inverse transformer 214, the decoded image signal S318 is generated as a result of performing the hybrid inverse transform, and the decoded image signal S318 is input to the motion compensation predictor 205, where the motion prediction of the hybrid transform coefficient signal S303 is performed. Will be

The adder 204 calculates the difference between the hybrid transform coefficient signal S303 of the original image signal S300 output from the block divider 202 and the hybrid transform coefficient signal S305 of the reference image output from the motion compensation predictor 205. I do.

The mode selector 203 outputs the hybrid transform coefficient signal S output from the block divider 202.
303 and the difference transform coefficient signal S304 output from the adder 204 are compared to select which one to encode for each block, and switches 215 and 216 are switch-controlled.

The quantizer 207 quantizes the signal output from the switch 215 to generate a quantized signal S31.
0 and sends it to the inverse quantizer 208. The inverse quantizer 208 inversely quantizes the quantized signal S310 to obtain an inverse quantized signal S311 and sends it to the adder 209.

The adder 209 restores the hybrid transform coefficient signal S312 by adding the inverse quantized signal S311 output from the inverse quantizer 208 and the signal S307 from the switch 216, and restores the resultant signal to the frame memory 21.
Send to 0.

The frame memory 210 stores the hybrid transform coefficient signal S312 restored by the adder 209 while rearranging the hybrid transform coefficient signal S312 into the form of a low-pass image (LL) and a high-pass image (LH, HL, HH). The inverse orthogonal transformer 211 outputs a low-frequency image (L) of the signal stored in the frame memory 210.
L) is subjected to inverse orthogonal transformation. The subband synthesizer 212 generates a reference image signal S318 by performing subband synthesis on the output of the inverse orthogonal transformer 211.

[0014]

By the way, the image coding apparatus shown in FIG. 25 performs sub-band division in the preceding stage,
In the subsequent stage, orthogonal transformation is performed, and only the low band image (LL) of the original image obtained by sub-band division is orthogonally transformed. For other high band images, the orthogonal transformation is not performed. Images are reconstructed by rearranging.

However, in the case of an image having many high-frequency components in the vertical and / or horizontal components, not only the low-frequency image but also the high-frequency image has a large energy. It is conceivable that high quality encoded and decoded images can be obtained by performing the process.

The present invention has been made in view of the above points, and is capable of obtaining high-quality encoded and decoded images with reduced block distortion even with a high compression ratio for various images. An encoding device, an image decoding device, and a method thereof are proposed.

[0017]

According to the present invention, an orthogonal transform coefficient is divided into a plurality of zones having different statistical properties, and the coefficients are individually encoded for each zone. As a result, the coding bit amount can be reduced.

Further, by using a plurality of trellis quantization means having a scalar quantization function, a quantization error can be reduced.

Further, by dividing an image into a plurality of band components, scanning each band image by dividing it into zones, and performing trellis quantization, compression encoding and decoding can be performed more easily and efficiently. Can be.

[0020]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below in detail with reference to the drawings.

(1) First Embodiment FIG. 1 shows an image coding apparatus ENC1 according to a first embodiment.
Is input to the orthogonal transformation unit 1. The orthogonal transform unit 1 converts the input image data D100 into a DCT (Discrete Cosine Transfo
rm) is performed, and orthogonal transform coefficient data D101 obtained as a result is sent to the quantization unit 2.

The quantizing section 2 quantizes the orthogonal transform coefficients to generate quantized coefficient data D102, and sends it to the zone division type scanning section 3. Zone division type scanning unit 3
Scans an input coefficient group for each of a plurality of divided zones.

That is, FIG. 2 shows that the input image is divided into blocks of 8 × 8 pixels BLK, DCT coefficients obtained by performing two-dimensional DCT on each block are further quantized, and the resulting quantized coefficients are horizontally And two-dimensionally arranged in a vertical direction from a low band to a high band. The zone division type scanning unit 3 shows a method of dividing this quantization coefficient into a direct current component (DC) and four zones (ZONE1, ZONE2, ZONE3 and ZONE4) to scan.

That is, the first zone ZONE1 is a coefficient group consisting of seven coefficients located at positions "1" to "7", and the second zone ZONE2 is located at positions "8" to "14". The third zone ZONE3 is a coefficient group composed of seven coefficients, and the third zone ZONE3 is located at positions "15" to "42".
Coefficient group, and a fourth zone Z
ONE4 is a coefficient group including 21 coefficients located at positions "43" to "63". Incidentally, the numbers assigned to each zone in FIG. 2 indicate the scanning order assigned to each quantization coefficient.

Accordingly, the 8 × 8 quantized coefficients are scanned in numerical order assigned to each of them, and are grouped for each zone and sent to the entropy coding unit 4 as scan output data D103. Entropy coding is performed for each zone.

As described above, the method of dividing the DCT coefficient (quantized coefficient) into a plurality of zones utilizes a point having a correlation between the characteristic of the image and the position of the orthogonal transform coefficient (DCT coefficient). It is. That is, when the orthogonal transform (DCT) is performed on the two-dimensional image, the power of the image component in the vertical direction appears as a coefficient value in the first zone ZONE1 shown in FIG. Zone Z
Appears as a coefficient value in ONE2.

Therefore, if the coefficient groups having different statistical properties are divided into zones and separately encoded, the information compression effect is enhanced. Therefore, the coefficients divided into zones as shown in FIG. 2 are encoded by the entropy encoding unit 4 (FIG. 1) for each zone. Here, a Huffman coding method is used as a coding method used in the entropy coding unit 4.

Incidentally, the coding method used as the entropy coding unit 4 uses, for example, arithmetic coding which shows a performance superior to Huffman coding for a memoryless information source, in addition to the Huffman coding method. You may do it.

The coefficients coded for each zone in the entropy coding unit 4 are output to the transmission system as a coded bit stream D104.

FIG. 3 shows an encoded bit stream D10.
4 shows an image decoding device DEC1 configured to decode the encoded bit stream D104. The encoded bit stream D104 output from the image encoding device ENC1 described above with reference to FIG.

The entropy decoding unit 5 inputs each of the coefficients arranged and encoded for each zone described above with reference to FIG. 2 as an encoded bit stream D104, and decodes the coefficients for each zone to obtain a quantized coefficient. Is restored, and the resulting quantized coefficient data D105 is sent to the zone division type reverse scanning unit 6. The zone division type reverse scanning unit 6 reversely scans the quantized coefficients arranged and decoded for each zone, thereby converting each of the 64 quantized coefficients into the data shown in FIG.
To the predetermined coefficient position.

The inversely scanned quantized coefficient data D106 is sent to the inverse quantizer 7 and subjected to inverse quantization processing, whereby the orthogonal transform coefficients (DCT coefficients) are restored. The orthogonal transform coefficient data D107 is supplied to the orthogonal inverse transform unit 8
Sent to The orthogonal inverse transform unit 8 performs an orthogonal inverse transform (inverse DCT) on the orthogonal transform coefficient to obtain the decoded image data D10.
8 is restored and output.

In the above configuration, the image encoding device EN
C1 is an image data of 8 × 8 (= 64) is converted into an 8 × 8 (= 64) orthogonal transform coefficient (DC) by the orthogonal transform unit 1.
(T coefficient) and further quantized by the quantization unit 2. The resulting 8 × 8 (= 64) quantized coefficients are read out in the scanning order as described above with reference to FIG. 2 and individually entropy-coded for each zone.

At this time, since each zone is grouped for each quantization coefficient having a similar statistical property, entropy coding is performed individually for each zone, so that a certain value in each zone is obtained. The frequency of occurrence increases. Therefore, when entropy coding is performed in which a short codeword is assigned to a value having a high frequency of occurrence, the compression efficiency increases for each zone.

Thus, according to the above configuration, the input image data D10
0 can be more efficiently compressed.

(2) Second Embodiment FIG. 4 in which parts corresponding to those in FIG.
Shows an image encoding device ENC2 according to the embodiment of the present invention, and converts the digitized input image signal D100 into an orthogonal transform unit 1.
To enter. The orthogonal transform unit 1 performs an orthogonal transform (two-dimensional DCT) using DCT on the input image signal D100, and sends the orthogonal transform coefficient data D101 obtained as a result to a trellis quantization unit 9.

The trellis quantization section 9 has a plurality of scalar quantization functions (D0, D1, D2 and D3), and is 8 × 8 (= 64) input as orthogonal transform coefficient data D101.
For each orthogonal transform coefficient (DCT coefficient) of the 63 pixels excluding the DC component (DC component) of the DCT block in pixel units, a plurality of scalar quantization functions (D ₀ , D ₁ , D ₂ and D ₃ ), the DC
A search is made as to whether the quantization error of the entire T block is minimized, and quantization is performed for each pixel using the scalar quantization function obtained thereby.

That is, FIG. 5 shows four scalar quantization functions D ₀ , D ₁ and D set in the trellis quantization section 9.
In the case of using ₂ and D _3, a representation on the number line quantized values that can take the respective scalar quantization function. Here, in FIG. 5, Δ is a basic unit, and an integer value (10 in this embodiment) is assigned. Accordance connexion, for example, when the scalar quantization function D ₀ is selected, a value corresponding to the scalar quantization function _{D 0 (......, -8Δ (=} - 80), - 4
Δ (= − 40), 0, 3Δ (= 30), 7Δ (= 7
0), 11Δ (= 110),...) Are possible values as quantization values. The trellis quantization unit 9 performs the orthogonal transform coefficient (DCT) to be quantized at this time among the quantized values that can be obtained by such a scalar quantization function (for example, D ₀ ).
The quantization value closest to the value of the coefficient is selected as the quantization value of the scalar quantization function.

In this way, the trellis quantization unit 9
For each scalar quantization function (D ₀ , D ₁ , D _2, and D ₃ ), of the possible quantization values, the quantization value closest to the orthogonal transform coefficient (DCT coefficient) to be quantized at this time Is obtained as a quantization value of each scalar quantization function.

Here, the trellis quantization unit 9 performs a plurality of scalar quantization functions (D) on each of the 63 orthogonal transform coefficients of the DCT block except for the DC component.
₀ , D ₁ , D ₂ and D ₃ ) are selected and quantized sequentially. At this time, a search is made for 63 combinations of quantization functions that minimize the sum of the quantization errors of all 63 coefficients.

That is, FIG. 6 shows a path for searching for a combination of scalar quantization functions (D ₀ , D ₁ , D ₂ or D ₃ ) that minimize the total quantization error for 63 orthogonal transform coefficients. The figure shows that coefficients 1 to 63 are orthogonal transform coefficients (DCT).
Coefficients) are arranged in the scanning order when zigzag scanning is performed from the low band to the high band.

Accordingly, for the coefficient 1 on the lowest frequency side excluding the DC component, first, of the two quantization functions set at the nodes NOD _{1,1 to} NOD _8,1 , the quantization is to be performed at this time. A scalar quantization function that can take a quantization value close to the coefficient 1 is selected. For example, assuming that the value of the coefficient 1 is 23, a scalar quantization function that can take a quantization value close to the coefficient value 23 is selected from the scalar quantization functions D _{0 and} D _{2 at} the node NOD _1,1 . . In this case, since Δ = 10 in FIG. 5, the scalar quantization function D ₀ can take “30” as the quantization value, whereas the scalar quantization function D ₂ uses “1” as the quantization value.
0 ". Therefore, when the scalar quantization function D ₀ is used, the quantization error is “7” for the value of the coefficient 1 “23”, and when the scalar quantization function D ₂ is used, the coefficient 1 The quantization error is “13” for the value “23” of the above.

Therefore, in this case, the scalar quantization function D ₀ is selected at the nodes NOD _1,1 . In this way, the scalar quantization function with the smaller quantization error is selected at each of the nodes NOD _{1,1 to} NOD _{8,1 and} the node of the coefficient 2 along the path corresponding to the selected quantization function. Turning to NOD _1,2 ~NOD _8,2. For example, when the upper scalar quantization function D ₀ is selected from the upper and lower scalar quantization functions D ₀ or D ₂ corresponding to the nodes NOD _1,1 in FIG. 6, the trellis quantization unit 9
Among paths proceeding from the node NOD _{1, 1,} the process proceeds to node NOD _{1, 2} for the coefficient 2 to select an upper path corresponding to the scalar quantization function D _0.

At this time, each node NOD
Difference (quantization error) between the possible quantization value of the scalar quantization function selected in _{1,1 to} NOD _8,1 and the value of coefficient 1
A node in the subsequent coefficient 2 NOD _1,2 ~NOD _8,2
In it is added to the quantization error of the selected scalar quantization function, further subsequent node NOD _{1, 3} of the coefficient 3 ~NOD _8,3
Is propagated to

Thus, when the selection of the scalar quantization function at each node for the 63 coefficients is completed, the accumulation of the quantization error for each coefficient of the selected scalar quantization function at this time is calculated at the end T1 of each path. Appears at ~ T8. Accordingly, the trellis quantization unit 9 selects the terminal (for example, T5) at which the smallest value appears from the accumulation of the plurality of quantization errors, and traces the path in which this terminal (T5) is obtained in the reverse direction. . The resulting path is a combination of scalar quantization functions for each coefficient that achieves the minimum quantization error.

Therefore, of the quantized values obtained when the corresponding coefficients are quantized by the combination of the searched scalar quantization functions, the coefficients (...,... -5, -4, -3, -2, -1, 0,
1, 2, 3, 4, 5, 6,...) Are sequentially transmitted to the subsequent scanning unit 10 as quantization coefficient data D102 (FIG. 4).

FIG. 7 shows a procedure for searching for a quantized value in the trellis quantizing section 9. That is, in FIG. 7, when the trellis quantization unit 9 enters the processing procedure from step SP0, the state number s is set to 1 in step SP1. The state number is a number corresponding to the position of the start point of the route in FIG. 6, and the state number s = 1 is the node NOD
_This shows the state starting from _1,1 .

Accordingly, the trellis quantizing section 9 performs step S
After setting the state number s to 1 in P1, the subsequent step S
In P2, the conversion coefficient number N = 1. This allows
The trellis quantization unit 9 enters a state of selecting one of two scalar quantization functions (in this case, D ₀ and D ₂ ) for the transform coefficient 1. Then, in the following step SP3, a quantization error by the scalar quantization function selected is calculated, and the quantization value at this time is stored.

Further, the trellis quantization section 9 performs the step S
After moving to P4, it moves to the next coefficient, and the conversion coefficient number N is 1
Is added, and it is determined in step SP5 whether the value of N at this time is 63 or more. If a negative result is obtained here, this means that the paths for all the coefficients (63) have not been obtained for the first state (s = 1), and at this time, the trellis quantization unit 9 The processing of steps SP3 and SP4 described above is repeated. On the other hand, if a positive result is obtained in step SP5, this means that the paths for all the coefficients have been found for the first state (s = 1), and at this time, the trellis quantization unit 9 The process proceeds to step SP6, where the process proceeds to the second state (s = 2). That is, the second
6, the node NOD for the coefficient 1 in FIG.
Indicates the state where route search starts from _2,1 .

In this way, the trellis quantization unit 9 repeats the above-described processing of steps SP2 to SP6 until the path search is performed for all the states (eight states in this embodiment). If a positive result is obtained in step SP7, this means that the path search has been completed for all the states. At this time, the trellis quantization unit 9 proceeds to step SP8, and all the states S = 1 to 8 Among them, a state in which the accumulation of the quantization error is minimized is detected.

When the state where the accumulation of the quantization error is minimized is detected in step SP8, the trellis quantization section 9
Moves to step SP9, and reversely detects the quantization path detected in step SP8 to detect the quantization value selected for each coefficient. In this way, the quantized value of each coefficient that minimizes the quantization error for one DCT block is obtained, and the trellis quantizing unit 9 performs each orthogonal transform on only the quantized coefficient obtained by removing the reference value Δ from the obtained quantized value. The coefficients are sent to the scanning unit 10, and the route search processing procedure ends in step SP10.

[0052] Here, each node NOD _ij in FIG. 6 is assigned a weighting factor W _ij to be multiplied by the quantization error. Focusing on the fact that the lower the orthogonal transformation coefficient is, the higher the energy is, the weight coefficient W _ij is calculated by the following equation:

[0053]

(Equation 1)

A value that satisfies the following is assigned. Incidentally, all of the weight coefficients may be "1".

Thus, the trellis quantization unit 9 selects a scalar quantization function that minimizes the accumulation of quantization errors for each DCT block in accordance with each orthogonal transform coefficient, and performs quantization. The resulting quantized coefficients are sent to the subsequent scanning unit 10 as quantized coefficient data D102 (FIG. 4).

The scanning section 10 performs zigzag scanning of the input quantized coefficients from the low band to the high band, and sends them to the entropy coding section 4 as scan output data D103. The entropy encoder 4 outputs the scan output data D1
By performing entropy coding on the quantized coefficient input as 03 using a method such as Huffman coding, a coded bit stream D104 is obtained, and this is output to the transmission path.

Incidentally, as the encoding method used as the entropy encoding unit 4, in addition to the Huffman encoding method, for example, arithmetic encoding or the like which shows a performance superior to Huffman encoding for a memoryless information source is used. You may do it.

FIG. 8 shows an image decoding device DEC2 adapted to decode the coded bit stream D104 output from the image coding device ENC2. The code output from the image coding device ENC2 described above with reference to FIG. The encoded bit stream D104 is converted to an entropy decoding unit 5
To enter.

The entropy decoding unit 5 restores the quantized coefficients by sequentially decoding the coded bit stream D104, and obtains the resulting quantized coefficient data D10.
5 is sent to the reverse scanning unit 11. The reverse scanning unit 11 is a DCT
By reverse-scanning the decoded quantized coefficients input to the image decoding device DEC2 in the zigzag scan order from the low band to the high band of the block, 64 quantized coefficients are returned to the original coefficient positions. return.

The quantized coefficient data D106 that has been inversely scanned in this way is sent to the trellis inverse quantization unit 12 and is subjected to trellis inverse quantization. In this trellis dequantization process, the trellis dequantization unit 12 outputs the quantized coefficient data D10
The orthogonal transform coefficient (DCT coefficient) corresponding to each quantized coefficient is restored by multiplying the quantized coefficient of each orthogonal transform coefficient input as 6 by the basic unit Δ. The orthogonal transform coefficient data D107 is sent to the orthogonal inverse transform unit 8. The orthogonal inverse transform unit 8 restores the decoded image data D108 by performing an orthogonal inverse transform (inverse DCT) on the orthogonal transform coefficient, and outputs the decoded image data D108.

In the above configuration, the image encoding device EN
C2 obtains a quantization coefficient that minimizes a quantization error by performing trellis quantization on orthogonal transform coefficients (DCT coefficients). In this case, the trellis quantization unit 9 calculates 1
One screen is divided into blocks of 8 × 8 pixels, and the above-described path search is performed on the orthogonally transformed coefficients obtained by the orthogonal transformation. For example, trellis quantization is performed on the coefficients obtained as a result of the wavelet transformation. , The number of handled coefficients is significantly reduced. Therefore, quantization can be performed more easily.

Each coefficient obtained in the orthogonal transform (DCT) has a characteristic that the energy becomes higher as the frequency becomes lower, so that the orthogonal transform coefficient is zigzag-scanned so that the orthogonal transform coefficient has a higher energy. It can be arranged in order from the low band to the high band with low energy. Therefore, by sequentially changing the weighted value to be multiplied at the node NOD _{ij with} respect to the arranged orthogonal transform coefficients, it is possible to perform quantization in accordance with the characteristics of the orthogonal transform coefficients.

Thus, according to the above configuration, by combining orthogonal transform (DCT) and trellis quantization,
Quantization can be performed more easily and with a smaller quantization error.

In the above-described second embodiment, the case where the scanning unit 10 that performs zigzag scanning on the quantized coefficients is used has been described. However, the present invention is not limited to this. The above-described zone division type scanning unit 3 may be used. In this way, the weight value W _ij can be set in accordance with the characteristic of the value that can be taken as a coefficient value for each zone.

In this case, in this case, the image decoding device DEC2
, The zone division type reverse scanning unit 6 described above with reference to FIG. 3 may be used.

(3) Third Embodiment FIG. 9 in which parts corresponding to those in FIG.
1 shows an image encoding device ENC3 according to an embodiment of the present invention, and inputs digitalized input image data D100 to a resolution conversion unit 13. The resolution conversion unit 13 uses the down-sample filter, for example, for the input image data D100 to reduce the resolution in the horizontal and vertical directions by 1 / N.
(N is an integer). Thus, the image data D109 whose resolution has been converted by the resolution conversion unit 13 is sent to the orthogonal conversion unit 1.

The orthogonal transform unit 1 receives the input image data D109
, And performs orthogonal transform (two-dimensional DCT) using DCT, and sends the resulting orthogonal transform coefficient data D101 to the quantization unit 2.

The quantizing section 2 quantizes the orthogonal transform coefficients to generate quantized coefficient data D 102, and sends it to the zone division type scanning section 3. Zone division type scanning unit 3
Scans the input coefficient group for each of the plurality of divided zones in the same manner as in the case described above with reference to FIG.

Accordingly, the block quantization coefficients of 8.times.8 pixels are scanned in numerical order assigned to each block, and are grouped for each zone to obtain the scan output data D1.
03 is sent to the entropy encoding unit 4 and entropy encoded for each zone. The coefficients coded for each zone in the entropy coding unit 4 are output to the transmission system as a coded bit stream D104.

FIG. 10 in which parts corresponding to those in FIG. 3 are assigned the same reference numerals as in FIG. 10 shows an image decoding apparatus DEC3 adapted to decode the coded bit stream D104 output from the image coding apparatus ENC3. Image coding device EN
The coded bit stream D104 output from C3 is input to the entropy decoding unit 5.

The entropy decoding unit 5 restores the quantized coefficients by decoding the coded bit stream D104 for each zone, and sends out the quantized coefficient data D105 obtained as a result to the zone division type reverse scanning unit 6. I do. The zone division type reverse scanning unit 6 reversely scans the quantized coefficients arranged and decoded for each zone, thereby returning each of the 64 quantized coefficients to a predetermined coefficient position shown in FIG.

The inversely scanned quantized coefficient data D106 is sent to the inverse quantizer 7 and subjected to inverse quantization processing, whereby the orthogonal transform coefficients (DCT coefficients) are restored. The orthogonal transform coefficient data D107 is supplied to the orthogonal inverse transform unit 8
Sent to The orthogonal inverse transform unit 8 restores the resolution-converted image data D108 by performing orthogonal inverse transform (inverse DCT) on the orthogonal transform coefficient, and sends it to the resolution inverse transform unit 14.

The resolution inverse converter 14 converts the image data D10
For example, the resolution is converted by the resolution conversion unit 13 of the image encoding device ENC3 (FIG. 9) by multiplying the resolution of N8 in the horizontal and vertical directions (N is an integer) by using, for example, an upsample filter. The previous input image data is restored, and this is output as restored image data D110.

In the above configuration, the image encoding device EN
C3 converts the resolution of the input image in the resolution conversion unit 13 so that, for example, when the input image is large, or the compressed image encoded by the image encoding device ENC3 is transmitted through a communication line with a narrow transmission bandwidth. In such a case, the resolution of the image can be reduced and transmitted.

Therefore, according to the above configuration, the compression efficiency of the input image data D100 can be further improved by combining the zone division type scanning section 3 and the resolution conversion section 13 to compress and encode the input image data D100.

(4) Fourth Embodiment FIG. 11 in which parts corresponding to those in FIG. 4 are assigned the same reference numerals as in FIG. 11 shows an image encoding device ENC4 according to the fourth embodiment.
The digitized input image data D100 is input to the resolution conversion unit 13. The resolution conversion unit 13 uses the down-sample filter, for example, for the input image data D100 to reduce the resolution in the horizontal and vertical directions by 1 / N.
(N is an integer). Thus, the image data D109 whose resolution has been converted by the resolution conversion unit 13 is sent to the orthogonal conversion unit 1.

The orthogonal transform unit 1 receives the input image data D109
, And performs orthogonal transform (two-dimensional DCT) using DCT, and sends orthogonal transform coefficient data D101 obtained as a result to the trellis quantization unit 9.

The trellis quantization unit 9 calculates the orthogonal transform coefficient as shown in FIG.
7 is generated by performing trellis quantization using the method described above with reference to FIG. 7, and is sent to the scanning unit 10. The scanning unit 10 converts the input quantization coefficient into the same as in the case described above with reference to FIG.
A zigzag scan is performed from the low band to the high band, and the zigzag scan is sent to the entropy encoding unit 4 as scan output data D103. The entropy coding unit 4 obtains a coded bit stream D104 by performing entropy coding on the quantized coefficient input as the scan output data D103 using a technique such as Huffman coding, and outputs this to the transmission path.

Incidentally, the coding method used as the entropy coding unit 4 uses, for example, arithmetic coding which shows a performance superior to Huffman coding for a memoryless information source, in addition to the Huffman coding method. You may do it.

FIG. 12 in which parts corresponding to those in FIG. 8 are assigned the same reference numerals, shows an image decoding device DEC4 adapted to decode the coded bit stream D104 output from the image coding device ENC4. Image coding device EN
The coded bit stream D104 output from C4 is input to the entropy decoding unit 5.

The entropy decoding unit 5 restores the quantized coefficients by sequentially decoding the coded bit stream D104, and obtains the resulting quantized coefficient data D10.
5 is sent to the reverse scanning unit 11. The reverse scanning unit 11 is a DCT
By reverse-scanning the decoded quantized coefficients input to the image decoding device DEC4 in the zigzag scan order from the low band to the high band of the block, 64 quantized coefficients are returned to the original coefficient positions. return.

The quantized coefficient data D106 that has been inversely scanned in this way is sent to the trellis inverse quantization unit 12 and subjected to trellis inverse quantization. In the trellis inverse quantization process, the trellis inverse quantization unit 12 multiplies the quantization coefficient of each orthogonal transform coefficient input as the quantization coefficient data D106 by the basic unit Δ in the same manner as described above with reference to FIG. Thus, the orthogonal transform coefficients (DCT coefficients) corresponding to the respective quantized coefficients are restored. This orthogonal transform coefficient data D
107 is sent to the orthogonal inverse transform unit 8. Inverse orthogonal transform unit 8
Restores the image data D108 in a state where the resolution has been converted by performing orthogonal inverse transform (inverse DCT) on the orthogonal transform coefficient, and sends it to the resolution inverse transform unit 14.

The resolution inverse converter 14 converts the image data D10
For example, the resolution of the image encoding device ENC4 (FIG. 11) is converted by multiplying the resolution by 8 (N is an integer) in the horizontal and vertical directions using an upsample filter, for example. The previous input image data is restored, and this is output as restored image data D110.

In the above configuration, the image encoding device EN
The C4 converts the resolution of the input image by the resolution conversion unit 13 so that, for example, when the input image is large, or when the compressed image encoded by the image encoding device ENC4 is transmitted via a communication line with a narrow transmission bandwidth. In such a case, the resolution of the image can be reduced and transmitted.

Therefore, according to the above configuration, the compression efficiency of the input image data D100 can be further increased by combining the trellis quantization section 9 and the resolution conversion section 13 to compress and encode the input image data D100.

(5) Fifth Embodiment FIG. 13 in which parts corresponding to those in FIG. 1 are assigned the same reference numerals as in FIG. 13 shows an image encoding device ENC5 according to a fifth embodiment.
The digitized input image data D100 is input to the band dividing unit 15. The band dividing unit 15 divides the input image data D100 into four bands (LL, HL, LH, and HH).

That is, FIG. 14 shows a band division state in the band division section 15, in which each screen of the input image data D100 is band-divided with respect to the horizontal frequency and the vertical frequency. The band (HH) component represents a high-frequency component image having both high horizontal and vertical frequencies, the band (HL) component represents a high-frequency component image having a high horizontal frequency and low vertical frequency, and the band (L
The H) component represents a high frequency component image having a low horizontal frequency and a high vertical frequency, and the band (LL) component represents a low frequency component image having both a low horizontal frequency and a low vertical frequency.

The low-frequency (LL) component image has four more bands (LLLL, LLHL, LLLH, and LLHH).
Divided into components. That is, the band (LLLL) component is
In the low-frequency component image, the horizontal frequency and the vertical frequency are both lower component images, and the band (LLHL) component is
The low frequency component image represents a component image having a relatively high horizontal frequency and a low vertical frequency, and the band (LLLLH) component is a component image having a low horizontal frequency and a relatively high vertical frequency among the low frequency component images. Represents the band (LLHH)
The component represents a component image in which both the horizontal frequency and the vertical frequency are relatively high in the low-frequency component image.

The band dividing section 15 outputs the low-frequency (LL) component image data D11
1, high frequency (HL) component image data D112, high frequency (L
H) The component image data D113 and the high frequency (HH) component image data D114 are sent to the orthogonal transform unit 1.

The orthogonal transform unit 1 performs orthogonal transform (two-dimensional DC) on each of the input component image data (D111, D112, D113 and D114) using DCT.
T), and the resulting orthogonal transform coefficient data D1
15, D116, D117 and D118 are sent to the quantization unit 2.

The quantizing section 2 calculates each orthogonal transform coefficient data D11
5, D116, D117 and D118 are quantized to obtain quantized coefficient data D119, D120 and D121.
And D122 are generated, and these are generated by the zone division type scanning unit 3.
To send to. The zone division type scanning unit 3 receives the input quantized coefficient data D119, D120, D121, and D1.
Scanning is performed for each of the zones 22 individually.

FIG. 15 shows the quantization coefficient data D1
19, D120, D121 and D122 are individually set for the low-frequency (LL) component image, the high-frequency (HL) component image, the high-frequency (LH) component image, and the high-frequency (HH) component image. Four zones (ZONE1-Z
ONE4). In FIG. 15, in the low-frequency (LL) component image, the energy distribution tends to be more deviated toward the low-frequency (upper left in FIG. 15) coefficient, whereas the high-frequency
In the L) component image, the energy distribution tends to be more deviated as the coefficient has a higher horizontal frequency (upper right in FIG. 15), and the coefficient in the higher frequency (LH) component image has a higher vertical frequency (FIG. 1).
5, the lower the left), the higher the energy distribution tends to be, and the higher the frequency (HH) component image, the higher the horizontal frequency and the higher the vertical frequency are (the lower right in FIG. 15), the more the energy distribution tends to be biased.

In each component image, the power of the image component in the vertical direction appears as a coefficient value in the first zone ZONE1 set in each component image, and the power of the image component in the horizontal direction is set in each component image. Appears in the second zone ZONE2 as a coefficient value. Thus, the probability that a coefficient of a certain value appears in each zone set in each component image is increased, so that the coefficient is scanned for each zone, and the coefficient is individually scanned for each component image. To send to.

Accordingly, the low-frequency (LL) component image scanning data D123 output by scanning the low-frequency (LL) component image for each zone and the high-frequency (HL) component image are scanned for each zone. High frequency (H
L) component image scanning data D124, high frequency (LH) component image scanning data D125 output by scanning the high frequency (LH) component image for each zone, and high frequency (H
The high-frequency (HH) component image scan data D126 output by scanning the H) component image for each zone is sent to the entropy coding unit 4.

The entropy coding unit 4 scans the output data (D123, D124,... D) for each component image in each band.
125 and D126), the quantization coefficients are entropy-coded for each zone. Thus, the quantization coefficients of each component image coded for each zone in the entropy coding unit 4 are the coded bit stream D
The multiplexed signal is output to the transmission system.

FIG. 16, in which parts corresponding to those in FIG. 3 are assigned the same reference numerals, shows an image decoding device DEC5 adapted to decode the coded bit stream D127 output from the image coding device ENC5. Image coding device EN
The coded bit stream D127 output from C5 is input to the entropy decoding unit 5.

The entropy decoding unit 5 restores the quantized coefficients for each band component multiplexed on the coded bit stream D127, and divides the resulting quantized coefficient data D123, D124, D125 and D126 into zones. It is sent to the pattern reverse scanning unit 6. Zone division type reverse scanning unit 6
Are the quantization coefficients (D123, D124, D125 and D125) for each band component arranged and decoded for each zone.
126) is inversely scanned for each band component, thereby returning each band component to the predetermined coefficient position shown in FIG.

The quantization coefficient data D119, D120, D121, and D
Reference numeral 122 denotes an orthogonal transform coefficient (DCT coefficient) for each band component, which is transmitted to the inverse quantization unit 7 and subjected to inverse quantization processing. The orthogonal transform coefficient data D128, D129, D130 and D for each band component
131 is sent to the orthogonal inverse transform unit 8. Inverse orthogonal transform unit 8
Is the orthogonal inverse transform (inverse DC) of the orthogonal transform coefficient for each band component.
T), the image data for each band component (low-band (LL) component image data D132, high-band (HL) component image data D133, and high-band (LH) component image data D13
4 and the high-frequency (HH) component image data D135), and sends them to the band synthesizing unit 16.

The band synthesizing section 16 synthesizes the image data (D132, D133, D134 and D135) of each band component to restore the input image data input to the image encoding section ENC5 (FIG. 13). This is output as restored image data D136.

In the above configuration, the image encoding device EN
C5 is input to the input image data D10 by the band division unit 15.
0 is divided into a plurality of band components. Then, the image of each of the band components (the low-frequency (LL) component image, the high-frequency (HL) component image, the high-frequency (LH) component image, and the high-frequency (HH) component image) obtained by dividing the band is divided into a plurality of zones. And scanning.

By dividing each band component in this way, the input image data is classified for each frequency-dependent feature as a feature of the image, and a zone having a high probability that a coefficient value appears for each band component is determined. By the division, entropy coding can be performed on a data group having a high probability that the same coefficient value is arranged (that is, each zone of each band component), and the compression efficiency can be further increased.

Thus, according to the above configuration, it is possible to increase the compression efficiency and realize high-efficiency encoding.

(6) Sixth Embodiment FIG. 17 in which parts corresponding to those in FIG. 4 are assigned the same reference numerals, shows an image coding apparatus ENC6 according to the sixth embodiment,
The digitized input image data D100 is input to the band dividing unit 15. The band dividing unit 15 divides the input image data D100 into four band components. These four band components are, similarly to the case described above with reference to FIG. 14, a low-frequency (LL) component image, a high-frequency (HL) component image, and a high-frequency (L
H) component image and high frequency (HH) component image.

The band dividing section 15 outputs low-frequency (LL) component image data D11
1, high frequency (HL) component image data D112, high frequency (L
H) The component image data D113 and the high frequency (HH) component image data D114 are sent to the orthogonal transform unit 1.

The orthogonal transform unit 1 performs orthogonal transform (two-dimensional DC) using DCT on each of the input component image data (D111, D112, D113, and D114).
T), and the resulting orthogonal transform coefficient data D1
15, D116, D117 and D118 are sent to the trellis quantization unit 9.

The trellis quantization unit 9 performs orthogonal transform coefficient data D115 and D116 obtained for each band component image.
By scanning D117 and D118 in the order shown in FIG. 18, for example, a coefficient group in the scanning order is obtained for each band component. The trellis quantization unit 9 arranges the coefficient groups thus obtained so as to form the paths described above with reference to FIG. 6, and performs trellis quantization on each band component image. The method of trellis quantization is the same as that described above with reference to FIGS. 5 to 7, and for each orthogonal transform coefficient of each band component, a quantization coefficient that minimizes the accumulation of quantization errors is obtained.

The trellis quantizing section 9 calculates the quantized coefficient data D119, D120, D1 obtained for each band component.
21 and D122 are sent to the scanning unit 10. Scanning unit 10
Are input quantized coefficient data D119, D12
0, D121, and D122 are zigzag-scanned from the low band to the high band in the same manner as described above with reference to FIG. 4, and the scan output data D123, D124, D
It is sent to the entropy encoder 4 as 125 and D126. The entropy encoder 4 outputs the scan output data D
The quantized coefficients input as 123, D124, D125 and D126 are entropy-encoded and multiplexed using a technique such as Huffman coding to obtain an encoded bit stream D127, which is output to a transmission path.

Incidentally, the coding method used as the entropy coding unit 4 uses, for example, arithmetic coding which shows a performance superior to Huffman coding for a memoryless information source, in addition to the Huffman coding method. You may do it.

Further, the scanning section 10 may be replaced with the zone division type scanning section 3 described above with reference to FIG. 13. In this case, the zone division as described above with reference to FIG. What is necessary is just to arrange a coefficient.

FIG. 19, in which parts corresponding to those in FIG. 8 are assigned the same reference numerals, shows an image decoding device DEC6 adapted to decode an encoded bit stream D127 output from the image encoding device ENC6. Image coding device EN
The coded bit stream D127 output from C6 is input to the entropy decoding unit 5.

The entropy decoding unit 5 restores the quantization coefficient for each band component by separately and sequentially decoding the data for each band component multiplexed in the coded bit stream D127. The quantized coefficient data D123, D124, D125 and D126 are sent to the reverse scanning unit 11. The inverse scanning unit 11 calculates the quantization coefficient of each band component image input to the image decoding device DEC6 and decoded in the zigzag scan order from the low band to the high band of the DCT block for each band component image. By performing reverse scanning, each quantized coefficient is returned to the original coefficient position.

The quantized coefficient data D 119, D 120, D 121 and D 122 thus inversely scanned are sent to the trellis inverse quantization section 12 and subjected to trellis inverse quantization.
In this trellis inverse quantization process, the trellis inverse quantization unit 12 performs the quantization coefficient data D119, D120, D121, and D in the same manner as described above with reference to FIG.
By multiplying the quantization coefficient of each orthogonal transformation coefficient input as 122 by the basic unit Δ, the orthogonal transformation coefficient (DCT coefficient) corresponding to each quantization coefficient is restored for each band component image. The orthogonal transform coefficient data D128, D12
9, D130 and D131 are sent to the orthogonal inverse transform unit 8. The orthogonal inverse transform unit 8 performs an orthogonal inverse transform (inverse DCT) of the orthogonal transform coefficient for each band component image, thereby obtaining image data (low-band (LL) component image data D13) for each band component.
2. High frequency (HL) component image data D133, high frequency (L
H) The component image data D134 and the high frequency (HH) component image data D135) are restored, and are sent to the band synthesizing unit 16.

The band synthesizing unit 16 restores the input image data input to the image encoding unit ENC6 (FIG. 17) by synthesizing the image data (D132, D133, D134 and D135) of each band component. This is output as restored image data D136.

In the above configuration, the image encoding device EN
By combining the band dividing unit 15 and the trellis quantization unit 9, the input image data is classified for each frequency-dependent feature as a feature of the image, and each of them is trellis-quantized separately. A quantization coefficient with a small quantization error is obtained for each band component image.

In this case, since each band component image is classified for each feature depending on the frequency, the probability that the same coefficient value appears is increased. Therefore, the compression efficiency can be increased by performing entropy coding.

Thus, according to the above configuration, by combining the band division unit 15 and the trellis quantization unit 9, it is possible to achieve an improvement in compression efficiency and a reduction in quantization error.

(7) Seventh Embodiment FIG. 20 in which parts corresponding to those in FIG.
17 shows a seventh embodiment of the image decoding device according to the present invention.
20, the image decoding device DEC7 has a post-filter unit 17 between the orthogonal inverse transform unit 8 and the band synthesizing unit 16.

The image data (D132, D133, D13) of each band component that has been orthogonally inverse-transformed by the orthogonal inverse transform unit 8
4 and D135) are subjected to predetermined filter processing in the post-filter unit 17, respectively. The post-filter processing includes, for example, a deblocking filter that suppresses block distortion that is conspicuous at a low bit rate when orthogonal transformation is performed in units of blocks (such as a DCT block of 8 × 8 pixels), or a strong edge unit. There is a de-ringing filter or the like that suppresses a component such as ringing that occurs to reduce image quality on a square.

Thus, the image data (D132, D133, D1) for each band component is
34 and D135) are filtered output image data D137, D138, D1
39 and D140 are sent to the band synthesizing unit 16, respectively.

The band synthesizing unit 16 outputs the filter output image data D137, D138, D13 of each input band component.
9 and D140, the original image is restored, and this is output as restored image data D141.

Thus, according to the above configuration, in the image decoding device DEC 7 having the zone division type scanning unit 6, the post-filter unit 17 is used to suppress block distortion, ringing, and the like for each band component image. The image quality of the restored image can be further improved.

(8) Eighth Embodiment FIG. 21 in which parts corresponding to those in FIG.
An eighth embodiment of the image decoding apparatus according to the present invention will be described.
In FIG. 21, the image decoding device DEC8 has a post-filter unit 17 between the orthogonal inverse transform unit 8 and the band synthesizing unit 16.

The image data (D132, D133, D13) of each band component orthogonally and inversely transformed by the orthogonal inverse transformation unit 8
4 and D135) are subjected to predetermined filter processing in the post-filter unit 17, respectively. The post-filter processing includes, for example, a deblocking filter that suppresses block distortion that is conspicuous at a low bit rate when orthogonal transformation is performed in units of blocks (such as a DCT block of 8 × 8 pixels), or a strong edge unit. There is a de-ringing filter or the like that suppresses a component such as ringing that occurs to reduce image quality on a square.

Thus, the image data (D132, D133, D1) for each band component is
34 and D135) are filtered output image data D137, D138, D1
39 and D140 are sent to the band synthesizing unit 16, respectively.

The band synthesizing unit 16 filters the output image data D137, D138, D13 of the input band components.
9 and D140, the original image is restored, and this is output as restored image data D141.

Thus, according to the above configuration, in the image decoding device DEC8 having the trellis inverse quantization unit 12,
By using the post-filter unit 17 to suppress block distortion, ringing, and the like for each band component image, the image quality of the restored image can be further improved.

(9) Ninth Embodiment FIG. 2 in which parts corresponding to those in FIG. 4 and FIG.
2 shows a ninth embodiment of the image encoding device according to the present invention. In FIG. 22, the image encoding device ENC9 includes:
Input image data D100 constituting a moving image is input to the subtracter 18. At this time, the motion compensation prediction unit 21 detects a motion vector between frames from the reference image stored in the image memory 20 and performs a motion compensation process using the motion vector to generate predicted image data D150. This is supplied to the subtractor 18.

The subtractor 18 calculates the difference between the input image data D100 and the predicted image data D150, and sends the result to the orthogonal transform unit 1 as predicted error image data D142. The orthogonal transform unit 1 performs orthogonal transform (DCT) on the prediction error image data D142 to obtain orthogonal transform coefficients, and sends the orthogonal transform coefficients to the trellis quantization unit 9 as orthogonal transform coefficient data D143.

The trellis quantization unit 9 quantizes the orthogonal transform coefficient data D143 into a quantization coefficient with a minimum quantization error by the method described above with reference to FIGS. The resulting quantized coefficient is quantized coefficient data D144.
As a zone division type scanning unit 3 and a trellis inverse quantization unit 1
2 is sent.

The zone division type scanning section 3 divides the quantized coefficient data D144 by the method described above with reference to FIG. 2, and performs each zone (ZONE1 to ZONE).
Scan the quantized coefficient data collectively for each 4). Thus, the quantized coefficient data for each zone is sent to the entropy encoding unit 4 as scan output data D145.
The entropy encoder 4 converts the scan output data D145 into
For example, an encoded bit stream D153 is obtained by performing entropy encoding using a technique such as Huffman encoding. This coded bit stream D152 is output to a predetermined transmission path.

The quantized coefficient data D144 output from the trellis quantization unit 9 is
And inversely quantized using the technique described above for FIG. The transform coefficient data D146 obtained as a result is input to the subsequent orthogonal inverse transform unit 8.

The orthogonal inverse transform unit 8 converts the transform coefficient data D14
6 by applying orthogonal inverse transform (inverse DCT) processing to generate restored image data D147 of the prediction error image,
This is sent to the adder 19. The adder 19 generates the decoded image data D148 by restoring the input image data by adding the predicted image data D150 output from the motion compensation prediction unit 21 to the restored image data D147. It is stored in the memory 20.

Incidentally, the motion vector detected by the motion compensation prediction unit 21 is transmitted to the image decoding device as motion vector data D151.

On the other hand, FIG.
Encoded bit stream D152 output from NC9
And a coded bit stream D152 input to the entropy decoding unit 5. The entropy decoding unit 5 performs an entropy decoding process on the coded bit stream D152, so that each of the zones (ZONE1-ZO) described above with reference to FIG.
NE4), restores the quantized coefficients scanned, and sends out the resulting quantized coefficient data D145 to the zone division type reverse scanning unit 6. The zone division type reverse scanning unit 6
By reverse-scanning the quantized coefficients arranged and decoded for each zone, the 64 quantized coefficients are returned to the predetermined coefficient positions shown in FIG.

The quantized coefficient data D144 inversely scanned in this manner is sent to the trellis inverse quantization unit 12 and subjected to trellis inverse quantization processing, thereby obtaining the orthogonal transform coefficients (DC).
T coefficient) is restored. This orthogonal transform coefficient data D14
6 is sent to the orthogonal inverse transform unit 8. The orthogonal inverse transform unit 8 restores the prediction error image data D147 by performing orthogonal inverse transform (inverse DCT) on the orthogonal transform coefficient, and outputs the restored image data D147.
To send to.

Here, the motion compensating unit 22 converts the predicted image data D149 stored in the image memory 20 into the motion vector data D151 transmitted from the image encoding device ENC9.
The motion compensation image data D150 is generated by performing motion compensation processing according to

Accordingly, the adder 19 generates restored image data D148 by adding the restored prediction error image data D147 and motion compensated image data D150.
This is output as a decoded image and the image memory 20
Is stored as a reference image for motion compensation of the next frame.

In the above configuration, the image encoding device ENC9 for encoding a moving image has orthogonal transform coefficients (DCT coefficients).
By performing trellis quantization on, a quantization coefficient that minimizes a quantization error is obtained. In this case, the trellis quantization unit 9 divides one screen into blocks of 8 × 8 pixels and performs the above-described path search with respect to the orthogonally transformed coefficients which have been orthogonally transformed, thereby obtaining, for example, a wavelet transform. Compared to the case where trellis quantization is performed on the coefficients obtained as a result, the number of handled coefficients is significantly reduced. Therefore, quantization can be performed more easily.

Each coefficient obtained in the orthogonal transform (DCT) has a characteristic that the energy becomes higher as the frequency becomes lower, so that the orthogonal transform coefficient is divided into four zones (ZONE1 to ZONE4) for scanning. By doing so (FIG. 2), those having the same value as the orthogonal transform coefficients can be arranged in one zone. Accordingly, by sequentially changing the weighted value to be multiplied for each of the arranged orthogonal transform coefficients for each node NOD _ij (FIG. 6) corresponding to the coefficient of each zone, it is possible to match the characteristics of the orthogonal transform coefficients. Quantization can be performed.

Thus, according to the above configuration, the combination of the orthogonal transform (DCT), the trellis quantization, and the zone division type scanning unit 3 makes it possible to perform quantization more easily and with a smaller quantization error. Can be realized.

(10) Other Embodiments In the above-described embodiment, the case where the division method shown in FIG. 2 is used in the zone division type scanning unit 3 has been described, but the present invention is not limited to this. For example, as shown in FIG. 24, four zones (ZONE1 to ZONE4) may be divided for each band. With this division, the zone ZONE1 where energy is concentrated,
In FIG. 6, it is easy to increase the weight W _ij to be multiplied at the time of trellis quantization described above.

Further, in the above-described embodiment, the case where eight states are used in the trellis quantization processing has been described. However, the present invention is not limited to this. You may make it variable.

The number of scalar quantization functions used in the trellis quantization processing is not limited to four (D ₀ , D ₁ , D ₂ and D ₃ ), and various other numbers of scalar quantization functions are used. Is also good.

Further, in the above-described embodiment, the case where DCT is used as the orthogonal transform means has been described. However, the present invention is not limited to this. For example, MDCT (Modified DC
T) or LOT (Lapped Orthogonal Transform) may be used.

[0145]

As described above, according to the present invention, the orthogonal transform coefficient is divided into a plurality of zones having different statistical properties.
By coding the coefficients individually for each zone, the amount of coding bits can be reduced.

Further, by using a plurality of trellis quantization means having a scalar quantization function, a quantization error can be reduced.

Further, by dividing an image into a plurality of band components, scanning each band image by dividing it into zones, and performing trellis quantization, compression encoding and decoding can be performed more easily and efficiently. Can be.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating an image encoding device according to a first embodiment of the present invention.

FIG. 2 is a schematic diagram used for describing a zone division type scanning method.

FIG. 3 is a block diagram showing an image decoding device according to the first embodiment of the present invention.

FIG. 4 is a block diagram showing an image coding apparatus according to a second embodiment of the present invention.

FIG. 5 is a schematic diagram illustrating quantization values that can be taken by a scalar quantization function.

FIG. 6 is a schematic diagram for describing trellis quantization.

FIG. 7 is a flowchart showing a trellis quantization processing procedure.

FIG. 8 is a block diagram illustrating an image decoding device according to a second embodiment of the present invention.

FIG. 9 is a block diagram showing an image encoding device according to a third embodiment of the present invention.

FIG. 10 is a block diagram showing an image decoding device according to a third embodiment of the present invention.

FIG. 11 is a block diagram showing an image encoding device according to a fourth embodiment of the present invention.

FIG. 12 is a block diagram showing an image decoding apparatus according to a fourth embodiment of the present invention.

FIG. 13 is a block diagram showing an image coding apparatus according to a fifth embodiment of the present invention.

FIG. 14 is a schematic diagram illustrating band division according to a fifth embodiment of the present invention.

FIG. 15 is a schematic diagram for explaining a zone division scanning method for each band.

FIG. 16 is a block diagram showing an image decoding device according to a fifth embodiment of the present invention.

FIG. 17 is a block diagram showing an image encoding device according to a sixth embodiment of the present invention.

FIG. 18 is a schematic diagram illustrating a coefficient scanning state for each band.

FIG. 19 is a block diagram showing an image decoding device according to a sixth embodiment of the present invention.

FIG. 20 is a block diagram showing an image decoding apparatus according to a seventh embodiment of the present invention.

FIG. 21 is a block diagram showing an image decoding apparatus according to an eighth embodiment of the present invention.

FIG. 22 is a block diagram illustrating an image encoding device according to a ninth embodiment of the present invention.

FIG. 23 is a block diagram showing an image decoding device according to a ninth embodiment of the present invention.

FIG. 24 is a schematic diagram illustrating a zone-type scanning method according to another embodiment.

FIG. 25 is a block diagram showing a configuration of a conventional image encoding device.

[Explanation of symbols]

1 ... Orthogonal transformation unit, 2 ... Quantization unit, 3 ... Zone division type scanning unit, 4 ... Entropy coding unit, 5 ... Entropy decoding unit, 6 ... Zone division type inverse scanning unit, 7 ......
Inverse quantization unit, 8: orthogonal inverse transformation unit, 9: trellis quantization unit, 10: scanning unit, 11: inverse scanning unit, 12: trellis inverse quantization unit, 13: resolution conversion unit, 14 resolution inversion section, 15 band division section, 16 band synthesis section, 17 post filter section, ENC1 to ENC9
... image encoding apparatus, DEC1~DEC9 ...... image decoding _{apparatus, D 0, D 1, D} 2, D 3 ...... quantization function, NOD _ij
……node.

Claims (42)

[Claims] An orthogonal transformation means for orthogonally transforming an input image; a quantization means for quantizing orthogonal transformation coefficients output from the orthogonal transformation means; Scanning means for dividing into a plurality of zones set according to the vertical spatial frequency and scanning each zone, and encoding means for encoding the quantization coefficient scanned by the scanning means. An image encoding device, characterized in that: 2. The image encoding apparatus according to claim 1, wherein said image encoding apparatus includes a resolution conversion unit for converting a resolution of said input image. 3. The image encoding apparatus according to claim 2, wherein the input image is a moving image, and the image encoding device sends a prediction error image, which is a difference between the input image and a predicted image predicted for the input image, to the orthogonal transform means. The image coding apparatus according to claim 1, wherein the input is performed. 4. An orthogonal transform coefficient is generated by orthogonally transforming an input image, a quantized coefficient is generated by quantizing the orthogonal transform coefficient, and the quantized coefficient is determined according to horizontal and vertical spatial frequencies. An image encoding method, wherein the image is divided into a plurality of set zones and scanned for each zone, and the scanned quantization coefficients are encoded. 5. The image encoding method according to claim 4, wherein, after converting the resolution of the input image, the orthogonal transformation is performed. 6. The image encoding method according to claim 6, wherein the input image is a moving image, and the image encoding method performs the orthogonal transform on a prediction error image that is a difference between the input image and a predicted image predicted for the input image. The image encoding method according to claim 4, wherein: 7. A decoding means for decoding an encoded bit stream and restoring a quantized coefficient, and separately decoding the restored quantized coefficient for each of predetermined zones according to horizontal and vertical spatial frequencies. Inverse scanning means for performing inverse scanning, inverse quantization means for inversely quantizing the quantized coefficient output by inverse scanning, and orthogonal inverse transform of the orthogonal transform coefficient restored by the inverse quantization means. An image decoding apparatus comprising: an orthogonal inverse transform unit. 8. The image decoding apparatus according to claim 7, wherein said image decoding apparatus includes a resolution inverse transforming means for inversely transforming the resolution of the image restored by said orthogonal inverse transforming means. . 9. The encoded bit stream is prediction error image data generated by encoding a moving image, and the image decoding device generates a motion compensated image based on motion vector information. The image decoding apparatus according to claim 7, further comprising a compensating unit, wherein the moving image is restored by adding the motion compensation image to a restored prediction error image output from the orthogonal inverse transform unit. 10. A decoded bit stream is decoded to restore a quantized coefficient, and the restored quantized coefficient is individually reverse-scanned for each of predetermined zones according to horizontal and vertical spatial frequencies. An image decoding method, comprising: inversely quantizing the quantized coefficient output after being inversely scanned and orthogonally inversely transforming the orthogonally transformed coefficient restored by the inverse quantization means. 11. The image decoding method according to claim 10, wherein in the image decoding method, the resolution of the image restored by the orthogonal inverse transformation is inversely transformed. 12. The encoded bit stream is prediction error image data generated by encoding a moving image. The image decoding method generates a motion compensated image based on motion vector information. 11. The image decoding method according to claim 10, wherein the moving image is restored by adding the motion compensation image to a restored prediction error image restored by the orthogonal inverse transform. 13. An orthogonal transformation means for outputting an orthogonal transformation coefficient based on the input image by orthogonally transforming the input image, and having a plurality of quantization functions, the orthogonal transformation coefficient being selected from the plurality of quantization functions. A quantization function that minimizes a quantization error with respect to the quantization function that quantizes the orthogonal transform coefficient using the selected quantization function; and a quantization output from the quantization means. An image encoding apparatus comprising: scanning means for scanning coefficients in a predetermined order; and encoding means for encoding the quantized coefficients scanned by the scanning means. 14. The image encoding apparatus according to claim 13, wherein said image encoding apparatus includes a resolution conversion unit for converting a resolution of said input image. 15. The quantization means quantizes the plurality of orthogonal transform coefficients generated by orthogonally transforming the input image with the plurality of quantization functions, respectively, and converts the plurality of orthogonal transform coefficients into the plurality of orthogonal transform coefficients. 14. The image encoding apparatus according to claim 13, wherein a combination of the quantization functions that minimizes the accumulation of quantization errors is determined. 16. The quantization means multiplies the quantization error by a weighting factor when quantizing each of the orthogonal transform coefficients by each of the plurality of quantization functions. Item 16. An image encoding device according to Item 15. 17. The image encoding apparatus according to claim 17, wherein the input image is a moving image, and the image encoding device sends a prediction error image, which is a difference between the input image and a predicted image predicted for the input image, to the orthogonal transform means. 14. The image encoding device according to claim 13, wherein the input is performed. 18. An orthogonal transformation of an input image to output an orthogonal transformation coefficient based on the input image, and a quantization function which minimizes a quantization error with respect to the orthogonal transformation coefficient from among a plurality of quantization functions. , Quantize the orthogonal transform coefficients by using the selected quantization function, scan the quantized coefficients generated by the quantization in a predetermined order, and scan the quantized coefficients. An image encoding method characterized by encoding. 19. The image encoding method according to claim 18, wherein the image encoding method converts a resolution of the input image. 20. The image encoding method, wherein the plurality of orthogonal transform coefficients generated by orthogonally transforming the input image are respectively quantized by the plurality of quantization functions, and the plurality of orthogonal transform coefficients are 19. The image encoding method according to claim 18, wherein a combination of the quantization functions that minimizes the accumulation of the quantization error is obtained for the image coding method. 21. The image encoding method, wherein each of the orthogonal transform coefficients is quantized by each of the plurality of quantization functions, wherein the quantization error is multiplied by a weight coefficient. The image encoding method according to claim 20. 22. The input image is a moving image, and the image encoding method performs the orthogonal transformation on a prediction error image that is a difference between the input image and a predicted image predicted for the input image. 19. The image encoding method according to claim 18, wherein: 23. A decoding means for decoding an encoded bit stream and restoring a quantized coefficient, an inverse scanning means for inversely scanning the restored quantized coefficient in a preset order, and A dequantizing means for dequantizing the quantized coefficient by multiplying the quantized coefficient outputted and output by a preset basic value; and an orthogonal transform coefficient restored by the dequantizing means. And an orthogonal inverse transform means for performing an orthogonal inverse transform on the image. 24. The image decoding apparatus according to claim 23, wherein said image decoding apparatus comprises a resolution inverse transforming means for inversely transforming the resolution of the image restored by said orthogonal inverse transforming means. . 25. The coded bit stream is prediction error image data generated by coding a moving image, and the image decoding device generates a motion compensated image based on motion vector information. 24. The image decoding apparatus according to claim 23, further comprising a compensation unit, wherein the moving image is restored by adding the motion compensation image to a restoration prediction error image output from the orthogonal inverse transform unit. 26. A coded bit stream is decoded to restore a quantized coefficient, the restored quantized coefficient is inversely scanned in a preset order, and the inversely scanned and output quantum is output. An image decoding method characterized in that the quantized coefficient is inversely quantized by multiplying the quantized coefficient by a preset basic value, and the orthogonal transform coefficient restored by the inverse quantization is subjected to orthogonal inverse transform. 27. The image decoding method according to claim 26, wherein in the image decoding method, the resolution of the image restored by the orthogonal inverse transformation is inversely transformed. 28. The encoded bit stream is prediction error image data generated by encoding a moving image. The image decoding method generates a motion compensated image based on motion vector information, 27. The image decoding method according to claim 26, wherein the moving image is restored by adding the motion compensation image to a restored prediction error image restored by the orthogonal inverse transform. 29. Band dividing means for dividing an input image into bands, orthogonal transform means for orthogonally transforming each of the band-divided images for each of the band components, and each of the bands outputted from the orthogonal transform means. A quantizing means for quantizing the orthogonal transform coefficients for each band component for each band component, and a quantizing coefficient for each band output from the quantizing means for each band component in a predetermined order. An image encoding apparatus comprising: a scanning unit that scans; and an encoding unit that encodes, for each band, the quantization coefficient of each band scanned by the scanning unit. 30. The scanning means, wherein each of the quantization coefficients for each of the bands output from the quantization means is divided into a plurality of zones set in accordance with horizontal and vertical spatial frequencies for each of the band components. 30. The scanning is performed for each of the band components separately.
An image encoding device according to claim 1. 31. The quantizing means includes a plurality of quantizing functions and, among the plurality of quantizing functions, a quantizing function which minimizes a quantization error with respect to the orthogonal transform coefficient. 30. The image coding apparatus according to claim 29, wherein the orthogonal transform coefficient is quantized for each of the band components using the selected quantization function. 32. An input image is band-divided, each of the band-divided images is subjected to orthogonal transform for each of the band components, and the orthogonal transform coefficient for each of the bands generated by the orthogonal transform is calculated. Each band component is quantized, and each quantized coefficient generated by the quantization for each band is scanned in a predetermined order for each band component. For each of the scanned bands, An image encoding method, wherein the quantization coefficient is encoded for each of the bands. 33. The image encoding method according to claim 28, wherein each of the quantization coefficients for each of the bands generated by the quantization is set for each of the band components in accordance with horizontal and vertical spatial frequencies. 33. The scanning is performed for each of the band components in each of the following zones.
2. The image encoding method according to 1., 34. The image encoding method, wherein a quantization function that minimizes a quantization error with respect to the orthogonal transform coefficient is selected for each of the band components from among a plurality of quantization functions. 4. The method according to claim 3, wherein the orthogonal transform coefficient is quantized for each of the band components using the quantization function.
3. The image encoding method according to item 2. 35. A decoding means for decoding an encoded bit stream to restore quantization coefficients for each of a plurality of band components, and decoding the restored quantization coefficients for each of the band components in a predetermined order. Inverse scanning means for performing inverse scanning for each of the bands, inverse quantization means for inversely quantizing the quantization coefficient for each of the band components output by inverse scanning for each of the band components, Orthogonal inverse transform means for orthogonally inverse transforming the orthogonal transform coefficient for each band component restored by the inverse quantization means for each band component, and each of the bands restored by the orthogonal inverse transform means An image decoding apparatus comprising: a band synthesizing unit that synthesizes an image for each component. 36. The inverse scanning means comprises: a plurality of quantized coefficients for each of the band components reconstructed by the decoding means which are set in accordance with horizontal and vertical spatial frequencies for each of the band components; 36. The image decoding apparatus according to claim 35, wherein reverse scanning is performed for each of the band components divided into zones. 37. The inverse quantization means multiplies the quantized coefficients by multiplying each of the quantized coefficients for each of the band components output by inverse scanning by a preset basic value. The image decoding device according to claim 35, wherein inverse quantization is performed for each band component. 38. The image decoding apparatus according to claim 31, further comprising: post-filter processing means for performing predetermined filter processing on the image for each band component restored by the orthogonal inverse transform means, wherein the filter processing is performed. 38. The image decoding device according to claim 35, wherein the restored image for each band is subjected to band combining by the band combining means. 39. An encoded bit stream is decoded to restore quantization coefficients for each of a plurality of band components, and the restored quantization coefficients for each of the band components are converted into the respective band components in a predetermined order. The inverse quantization is performed for each of the band components, and the quantization coefficient of each of the band components output after being inversely scanned is inversely quantized for each of the band components, and each of the bands restored by the inverse quantization means. An image decoding method, comprising: performing orthogonal inverse transform on an orthogonal transform coefficient for each component for each band component; and performing band synthesis on the image for each band component restored by the orthogonal inverse transform. 40. The image decoding method according to claim 27, wherein the quantized coefficients for each of the band components reconstructed by the decoding include a plurality of zones set for each of the band components in accordance with horizontal and vertical spatial frequencies, respectively. 4. A reverse scanning is performed for each of the band components separately.
10. The image decoding method according to 9. 41. The image decoding method, comprising: multiplying each of the quantized coefficients for each of the band components output by inverse scanning by a preset basic value to thereby convert the quantized coefficients to each of the band components. The image decoding method according to claim 39, wherein inverse quantization is performed for each component. 42. The image decoding method according to claim 42, wherein a predetermined filtering process is performed on the image for each of the band components restored by the orthogonal inverse transform, and each of the bands on which the filtering process has been performed is performed. 41. The restored image is subjected to the band synthesis.
Or the image decoding method according to 41.