JPH047989A - Picture signal encoding control system - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Summary] Regarding an image signal encoding control method for highly efficient encoding of image signals in video conferences, video telephones, etc., code assignment that takes into account playback image quality as well as improving encoding efficiency. A discrete cosine transform unit that performs a discrete cosine transform on an input image signal or an inter-frame difference signal of the input image signal, and a transform coefficient from the discrete cosine transform unit or a transform coefficient of the transform coefficient. comprising a quantization unit that quantizes the inter-frame difference signal, an encoding unit that performs variable length encoding on the quantized output signal from the quantization unit, and an image determination unit that determines the nature of the input image signal, By the image determination section,
Corresponding to blocks consisting of a plurality of pixels of the input image signal,
It is configured to determine whether it corresponds to a flat portion, an edge portion, or a fine pattern portion, and to control the quantization step of the quantization unit based on the determination result.

[Industrial Application Field] The present invention relates to an image signal encoding control method for highly efficient encoding of image signals in video conferences, video telephones, and the like.

Various high-efficiency coding methods for image signals have already been proposed, and recently one of them is the discrete cosine transform (DCT).
sforra) is attracting attention. This discrete cosine transformation uses multiple pixels as one block and orthogonally transforms them into transformation coefficients using a transformation matrix.It varies depending on the nature of the image, but - for boat fishing, the transformation coefficients are concentrated in the low frequency region. By quantizing the transform coefficients and performing variable length encoding, the image signal is highly efficiently encoded, and this encoded signal is transmitted or recorded in an image recording device. Then, at the time of reproduction from the receiving side or an image recording device, variable length decoding, inverse quantization, and further inverse discrete cosine transformation are performed to obtain a reproduced image signal.

It is desired to further improve the encoding efficiency in such a method of performing discrete cosine transform and encoding on a human image signal.

[Prior Art] FIG. 15 is a block diagram of a conventional example, in which a transmitting section includes a discrete cosine transform section 60, a quantizing section 61, a scanning section 62, and an encoding section 63, and a receiving section includes a decoding section. It includes a section 64, an inverse scanning section 65, an inverse quantization section 66, and an inverse discrete cosine transform section 67.

The input image signal is an image signal obtained by capturing a still image such as a moving image or a drawing with a television camera, an image signal reproduced from an image recording device, or an image signal subjected to frame difference processing, etc. This signal is converted into a digital signal by an AD converter (not shown), and is applied to the discrete cosine transform section 60. In this discrete cosine transform unit 60, for example, as shown in FIG. 16, one screen is divided into blocks each consisting of NXN pixels, and the pixels f (u, v) of each block are subjected to discrete cosine transform, The transform coefficients F(i, j) are individually quantized in the quantization unit 61, and the quantized output signal in the two-dimensional array is converted into one-dimensional by scanning in the horizontal direction, vertical direction, or zigzag direction in the scanning unit 62. The encoder 63 performs entropy encoding using variable-length encoding, etc., which assigns short codes to those with a high probability of occurrence and long codes to those with a low probability of occurrence.

On the receiving side, a decoding unit 64 performs entropy decoding on the received code in a process opposite to that of the encoding unit 63, an inverse scanning unit 65 performs scanning process inverse to that of the scanning unit 62 on the transmitting side, and an inverse quantization unit In step 66, the process opposite to that of the quantization unit 61 is performed, and in the inverse discrete cosine transform unit 67, the process is performed in the discrete cosine transform unit 6.
The image from the transmitting side is displayed by performing processing inversely to 0 to obtain an output image signal and adding it to a display device (not shown).

The discrete cosine transform is expressed by the following equations (1) and (2).

1+J=0+1+2+ ・ ・ ・N-1u, v-0,
L2. − − − N−1 Based on equations (1) and (2), the pixels f (u, ν) of each block in one screen
When converted into a conversion coefficient F(i, j), it indicates a frequency component corresponding to the value of i, j, and the smaller the value of i, j, the lower the frequency component. F(0,0) indicates the DC component. For example, if N=8 in a block consisting of N×N pixels on the screen in FIG.
The direction is the conversion coefficient for the low frequency region, and the direction indicated by the arrow is the conversion coefficient for the high frequency region. In the case of a -fishing screen, by performing discrete cosine transformation, the value of the transform coefficient for the low frequency component is large, the value of the transform coefficient for the high frequency component is small, and the value of the transform coefficient for the high frequency component becomes 0 in many cases.

This conversion coefficient F(i, j) is quantized in a quantization unit 61 and scanned in a scanning unit 62.
In the lower block of the figure, C is a zigzag scan, d(
Solid line) indicates horizontal scanning, e (-dotted chain line) indicates vertical scanning, and either of these scans scans from the low frequency region to the high frequency region or in the opposite direction,
A two-dimensional array is converted into a one-dimensional array, and the encoding unit 63 sequentially performs entropy encoding.

As mentioned above, the high frequency components of the conversion coefficients are often 0, so scanning is performed from the high frequency region side and the high frequency components are 0.
An end-of-bounds code EOB is added to the position where it no longer exists. For example, for the quantized signal of the 8 x 8 pixel block shown in Fig. 17, zigzag scanning is performed as shown by the arrow from the high frequency region side, and the position where the value is no longer 0 (-7 value position: upper left pixel position) is 1 and the lower right pixel position is 64, the end-of-bounds code EOB is added to the pixel position "46").

Then, when performing zigzag scanning from the low frequency region side, 10
31 (DC), -9, -5, 3, 2015, O,・
・The conversion count of the -dimensional array of O, -7 (EOB) is obtained, and it is entropy encoded in the encoding unit 63, and since all the transformation coefficients after the end-of-bound code EOB are 0, the encoding is omitted, thereby improving encoding efficiency.

On the receiving side, this end-of-pound code EOB
By detecting this, it is possible to identify the end of the encoded signal of the block and perform decoding.

[Problem B to be solved by the invention] By performing discrete cosine transform on the input image signal block-by-block, transform coefficients concentrated in the low frequency region are obtained, and O increases in the high frequency region. From this, O after the end-of-bounds code EOB on the high frequency region side
The encoding efficiency is improved by not transmitting the data.

However, in the conventional method of encoding image signals using discrete cosine transform, the images are uniformly encoded with high efficiency and then transmitted or recorded, regardless of the characteristics of the image, and visual characteristics are not taken into consideration. Since it does not perform high-efficiency encoding, it has the disadvantage that when the encoding efficiency is simply further improved, the reproduced image quality deteriorates.

It is an object of the present invention to improve coding efficiency and to enable code assignment that takes playback image quality into consideration.

[Means to solve the problem]

The image signal encoding control method of the present invention controls the quantization step by determining the properties of the input image signal, and will be explained with reference to FIG.

Discrete cosine transform unit 1 that performs discrete cosine transform on an input image signal or an inter-frame difference signal of the input image signal
and a quantization unit 2 that quantizes the transform coefficients from this discrete cosine transform unit (2) or the interframe difference signal of this transform coefficient.
, an encoding section 3 that performs variable length encoding on the quantized output signal from the quantization section 2, and an image determination section 4 that determines the nature of the input image signal. It determines whether a block consisting of a plurality of pixels of an image signal corresponds to a flat portion, an edge portion, or a fine pattern portion, and controls the quantization step of the quantization unit 2 based on the determination result. .

Furthermore, it includes a motion compensation section that performs motion compensation for blocks between frames of the input image signal, and the quantization section 2 is controlled by the image determination section 4.

Also, when the image determining section 4 determines that the area is flat,
The quantization step of the quantization unit 2 is made finer, and when it is judged to be an edge part, it is made coarser than the quantization step for a flat part, and when it is judged to be a fine pattern part, it is made coarser than the quantization step for a flat part. The quantization step is set to be the same as or coarser than the quantization step in the case of .

Further, the image determining unit 4 determines that the block is a flat portion when the standard deviation of the pixel values is equal to or less than the first threshold, and
When it exceeds a threshold value and is less than or equal to a second threshold value, it is determined to be an edge portion, and when it exceeds the second threshold value, it is determined to be a fine pattern portion.

In addition, the image determination unit 4 applies a Laplacian mask to the pixels of the block, and as a result, when the number of pixels exceeding a predetermined threshold is less than or equal to a third threshold, it is determined that the block is a flat portion. When it exceeds a threshold and is below a fourth threshold, it is determined to be an edge portion, and when it exceeds this fourth threshold, it is determined to be a fine pattern portion.

The image determination unit 4 also calculates the average value of the pixel values of the block or subblocks obtained by subdividing this block, and determines whether the absolute value of the difference between this average value and the pixel value within the block or subblock is a predetermined value. The number of pixels exceeding the threshold is counted, and when the number of pixels is less than or equal to a fifth threshold, it is determined to be a flat portion, and when the number of pixels exceeds the fifth threshold and is less than or equal to a sixth threshold, it is determined to be an edge portion, Further, when the number of pixels exceeds the sixth threshold value, it is determined that the pattern is a fine pattern part.

[Effect]

A discrete cosine transform unit 1 performs a discrete cosine transform on an input image signal or a discrete cosine transform on an inter-frame difference signal, and a quantizer 2 quantizes transform coefficients. The quantization unit 2 has a configuration for switching quantization steps or a configuration for switching quantizers of a plurality of different quantization steps, and the switching is controlled by the image determination unit 4. Further, the encoding unit 3 performs variable length encoding such as Huffman coding on the quantized output signal.

The image determining unit 4 determines whether each block of the input image signal corresponds to a flat part such as the sky or a wall, an edge part such as the outline of an object, or a fine pattern part such as a pattern. This is to switch the quantization step of the quantization section 2.

Further, the motion compensation unit detects a motion vector between the block of the input image signal of the current frame and the block of the input image signal of the previous frame, and calculates the interframe difference for the block with large motion in image body No. 2. This prevents the signal from increasing.

In addition, the visual characteristics have good sensitivity to flat areas and edge areas (relatively low sensitivity to fine pattern areas,
Even if there is some distortion, it cannot be identified. In addition, if the quantization step in the quantization unit 2 is coarse (this will result in a blurred reproduced image, and if the high frequency components of the transform coefficients obtained by performing discrete cosine transform are forced to zero, block distortion will occur). Therefore, the image determination unit 4 has a tendency to
Judgment is made for each block of the input image signal, and for flat areas, the quantization step is finer, for edge areas, the quantization step is the same as or coarser than the flat area (for fine pattern areas, the quantization step is finer). This is set to be similar to or coarser than the edge portion.Thereby, encoding efficiency can be improved without deteriorating the reproduced image quality.

In addition, the image determining unit 4 determines whether the input image signal corresponds to a flat portion, an edge portion, or a fine pattern portion based on the values (luminance) of pixels constituting the block. One way to do this is to first find the standard deviation of the pixel values of a block of input image signals.

Regarding this standard deviation, 1. Compared to the second threshold, when it is below the first threshold, it is composed of relatively average pixels, so when it is a flat part such as the sky or a wall, and when it exceeds the second threshold, it is a different value. Therefore, in a fine pattern part such as a striped pattern, when it exceeds the first threshold value and is less than the second threshold value, the pixel structure is intermediate between a flat part and a fine pattern part, Since it includes an outline part of a character 1 figure, etc., it is determined to be an edge part.

Another means in the image determination section 4 applies a Laplacian mask to the pixels of the block of the input image signal.

As a result, the number of pixels exceeding a predetermined threshold is detected and counted. That is, the number of pixels corresponding to edges is counted. When the number of pixels is below the third threshold, there are few pixels corresponding to edges, so it is a flat area, and when it exceeds the fourth threshold, there are many pixels corresponding to edges, so it is a fine pattern area. If it is below the fourth threshold, it is determined to be an edge portion even though the pixel configuration is intermediate between a flat portion and a fine pattern portion.

Further, another means in the image determining section 4 calculates the average value of the pixel values of the block or sub-block. Then, it is determined whether the absolute value of the difference between this average value and the value of each pixel exceeds a predetermined threshold, and the number of pixels exceeding the threshold is counted. When this number of pixels is less than or equal to the fifth threshold, the difference in pixel values of the block is small, so the block is determined to be a flat portion.

When the number of pixels exceeds the sixth threshold, the difference in pixel values in that block is large, so it is determined that the block has a fine pattern. When the number of pixels exceeds the fifth threshold and is less than or equal to the sixth threshold, the difference in pixel values of the block is larger in the flat part and smaller than in the fine pattern part, so the block is determined to be an edge part.

〔Example〕

Embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 2 is a block diagram of an embodiment of the present invention, and the transmission side includes a discrete cosine transform section 11, a quantization section I2, a scanning section 15, an encoding section 3, a transmission section 16, The image determination section 14 is also provided. The receiving side includes a receiving section 17, a decoding section 18, an inverse scanning section 19, an inverse quantizing section 20, and an inverse discrete cosine transform section 21.

A discrete cosine transform unit 11 performs discrete cosine transform on an input image signal such as a moving image signal from a television camera or the like in a block-by-block manner, a quantizer 12 quantizes the transform coefficients, and a scanning unit 15 In this step, the two-dimensional array corresponding to blocks is converted into a one-dimensional array by zigzag scanning, horizontal scanning, vertical scanning, etc., and the encoder 13! At this point, variable length coding is performed, and the transmitter 16 performs modulation, etc., and transmits the coded signal to the receiving side.

In addition, the image determining section 14 determines whether the input image signal corresponds to a block, whether it is a flat portion, an edge portion, or a fine pattern portion, and adds the determination signal to the quantizing section 12 and the transmitting section 16,
It controls the quantization step in the quantization section 12 and sends the quantization step as auxiliary information from the transmission section 16 to the receiving side. In this case, as in the case of a normal entropy coding transmission method, the amount of information to be transmitted is determined from the usage amount of the transmission buffer, and the quantization unit 12 determines the amount of information to be transmitted from the usage amount of the transmission buffer, and
It is also possible to add a configuration to control.

On the receiving side, the receiving section 17 receives the encoded signal,
The decoding unit 18 decodes the signal, the inverse scanning unit 19 converts the one-dimensional array signal into a two-dimensional array signal corresponding to blocks, and the inverse quantization unit 20 dequantizes the signal. At this time,
The receiving section 17 adds the auxiliary information indicating the separated quantization step to the quantization section 20, and inverse quantization is performed using the same quantization step as on the transmitting side. The inverse discrete cosine transform unit 21 performs inverse discrete cosine transform on the inverse quantized output signal to obtain a received image signal, which is displayed on a display device (not shown).

The image determination unit 14 determines the size of the number of pixels that is equal to or greater than a threshold value of the difference from the average value for each block of the input image signal.

Based on the standard deviation, the number of edge pixels, etc., it is determined whether the area is a flat area, an edge area, or a fine pattern area. Based on this determination signal, the quantization step of the quantization unit 12 is made finer for flat portions, coarser than the flat portion for edge portions, and coarser for edge portions. It is the same or rougher.

FIG. 3 is a block diagram of another embodiment of the present invention, 31
is a discrete cosine transform unit (DCT), and 32 is a quantization unit (
Q), 33 is a coding unit (VLC), 34 is an image judgment unit,
35 is an adder circuit, 36 is a high frequency component deletion circuit (HFC)
, 37 is an inverse quantization unit (Q), 38 is an addition circuit, 39 is a frame memory (FM), and 40 is a switching unit for switching between interframe coding and intraframe coding.

This embodiment shows a case where inter-frame differences are determined and encoded in the transform coefficient domain, and a discrete cosine transform unit 31 performs a discrete cosine transform on an input image signal, and the resulting transform coefficients and the resulting data are stored in a frame memory 39. The transform coefficients of the previous frame are added to the adder circuit 35 to obtain a difference, quantized by the quantizer 32, the transform coefficients of high frequency components are deleted by the high frequency component delete circuit 36, and the encoder 33 encodes the transform coefficients with variable length. and send it to the receiving side. Alternatively, it is recorded in addition to a direct image recording device.

Furthermore, the transform coefficients from which high frequency components have been removed by the high frequency component deletion circuit 36 are dequantized by the dequantizer 37, and added to the transform coefficients of the previous frame stored in the frame memory 39 in the adder circuit 38, and the current is stored in the frame memory 39 as a conversion coefficient for the frame.

Further, the amount of generated information between interframe encoding and intraframe encoding is compared, and when the amount of information generated in the case of intraframe encoding is small, the switching unit 40 is controlled to switch from the frame memory 39 side to the ground side. This allows intraframe encoding to be performed.

The image determination unit 34 also blocks the input image signal by
This is for determining whether it is a flat portion, an edge portion, or a fine pattern portion, and the quantization steps of the quantization unit 32 and the inverse quantization unit 37 are switched based on the determination signal.

Further, the high frequency component deletion circuit 36 performs, for example, zigzag scanning C1 horizontal scanning d for the two-dimensional array of blocks in FIG.
, vertical scan e, a predetermined number of low frequency components from the DC component, for example, (NxN)/2 including the DC component.
By extracting the quantized output signals of the transform coefficients and adding them to the encoding unit 33 after removing high frequency components,
Transform coefficients of high-frequency components having values of zero or values close to zero are not encoded, and thereby the encoding efficiency can also be improved. Note that the high frequency component deletion circuit 36 can also be provided before the quantization section 32.

Similarly to the above-described embodiment, the image determining section 34 determines whether the input image signal is a flat portion, an edge portion, or a fine pattern portion for each block, and then sends the quantization section 32 and the inverse quantization section 37.
This controls the quantization step.

The functions of each section such as the discrete cosine transform section 31 described above can be realized by respective dedicated integrated circuits or by the arithmetic functions of a digital signal processor.

FIG. 4 is a block diagram of still another embodiment of the present invention,
The case where motion compensation is performed is shown. In the figure, 41 is a discrete cosine transform unit (DCT), 42 is a quantization unit (Q), and 4
3 is a coding unit (VLC), 44 is an image determination unit, 45 is an addition circuit, 46 is a high frequency component deletion circuit (HFC), 47 is an inverse quantization unit (Q), 48 is an addition circuit, and 49 is a frame memory ( FM), 50 is a switching unit, 51 is an inverse discrete cosine transform unit (JDCT), and 52 is a motion compensation unit.

The input image signal is sent to an image determination section 44, an addition circuit 45,
The motion compensation section 52 is also added to the motion compensation section 52. When the switching unit 50 is switched to the motion compensation unit 52 side as shown, the input image signal and the image signal in the previous frame of the frame memory 49 are compared in block correspondence to detect a motion vector,
Control is performed so that the difference between frames becomes small. When the switching unit 50 is switched to the ground side, intra-frame encoding is performed, and when it is switched to the frame memory 49 side, inter-frame encoding without motion compensation is performed. The switching control by the switching unit 50 can also be performed by comparing the amount of information generated by each encoding method and selecting the encoding method that generates the least amount of information.

The difference between the input image signal and the image signal of the previous frame is determined by the addition circuit 45, and this difference signal is subjected to discrete cosine transformation by the discrete cosine transformation unit 41, quantized by the quantization unit 42, and then processed by the high frequency component deletion circuit. The transform coefficients of the high frequency components quantized by 46 are deleted, and variable length coding is performed in the coding unit 43, and the coded signal including auxiliary information such as quantization steps and motion vectors is not shown. The image is sent to the receiving side or recorded on an image recording device.

The transform coefficients dequantized by the dequantization unit 47 are subjected to inverse discrete cosine processing by the inverse discrete cosine transformation unit 51, and added to the image signal of the previous frame by the addition circuit 48, as the image signal of the current frame. It is stored in the frame memory 49.

Further, the image determining section 44 determines whether the input image signal is a flat portion, an edge portion, or a fine pattern portion corresponding to each block, and controls the quantization steps of the quantizing portion 42 and the inverse quantizing portion 47.

In this embodiment, when a moving image signal with large motion is input, the encoding efficiency can be improved by motion compensation, and the characteristics of the image in that case are determined and the quality of the reproduced image deteriorates. It is possible to control the quantization step so that it does not occur.

5 and 6 show flowcharts of image judgment based on standard deviation in the image judgment section. First, F (8,
As shown in 8), the block has a size of 8 x 8 pixels,
Let the standard deviation be σ, the first . Let the second threshold be THI, TH2, and the average value H. First, let the average value H be O(a2), then I = 1 (a3) and J = 1 (a4) (first pixel position of the block ), then the average value H=H+f
Find (I, J) (a5), then set J=J+1 and (
a6), determine whether J>8 or not; a7); if J>8, proceed to step (a5); if J>8, I=
Assuming I+1, a8), determine whether I > 8 or not, a9)
, if I>8, proceed to step (a4) and I>
In the case of 8, since the accumulation process in the 8×8 block has been completed, the average value H is determined by setting H=H/64 (alO).

Next, let σ = 0 and I = 1 (al2
) and J = 1 (al3), the standard deviation σ is
−σ+(f(I,J)−H)” (al4)
, that is, the square of the difference between the pixel value and the average value is accumulated. Then, J=J+1 is determined (al5), J > 8 is determined (al6), and if J>8 is not, the process moves to step (al4), and if J>8, I is determined to be I=I+1 (al7)). , determine whether 1>8 or not al8) , l
> If not 8, move to step (al3) and I>
In the case of 8, it means that the cumulative value within the block of 64 pixels has been obtained, so σ=177TT al9), and this is set as the standard value σ.

Then, it is determined whether σ<THI or not (a20). If the standard deviation σ is smaller than the first threshold THI, the difference in pixel values within the block is small, so it is determined that the block is a flat area such as the sky or a wall. do. Also, if the standard deviation σ is not less than the first threshold THI, it is determined whether THI≦σ≦TH2 or not.C21
), if the standard deviation σ exceeds the first threshold THL and is less than or equal to the second threshold TH2H2, there is some variation in the pixel values within the block, so the edge portion including the rounded portions of characters and figures, etc. judge. Also, the standard deviation σ is the second threshold T
If H2 is exceeded, there are many variations in pixel values within the block, so it is determined that the pattern is a fine pattern such as a striped pattern.

FIGS. 7 and 8 are flowcharts of image judgment based on edge detection, and show a case where edge detection is performed using Laplacian. That is, a Laplacian mask M with a size of 3 x 3 and a block f with a size of 8 x 8 pixels (8, 8)
, the block f' after applying the Laplacian mask M
(8,8), the threshold TH and the third . Fourth threshold TH3,
Define TH4 and the number of edges C0UNT (bl), I
= 2 (b2), J = 2 (b3). That is, a Laplacian mask M is applied to pixels excluding peripheral pixels of one block. In addition, step (bl
), the Laplacian mask M sets the center pixel to "4".
'', the pixels on the top, bottom, left and right sides are set as r-IJ, and the pixels in the diagonal top and bottom directions are set as "0", but it is also possible to use a mask other than this configuration.

And I=2. The pixel f (2, 2) of J=2 is subjected to the calculation shown in step (b4) and mask M is applied.Next, J=J+1 is set b5), and it is determined whether J>7 or not b6 ), if J>7, step (b
4), if J>7, I=I11 and bl
), determine whether I > 7 or not (b8), if I > 7, move to step (b3), if I > 7, calculate the Laplacian for 6 × 6 pixels in the 8 × 8 pixel block. Now that mask M has been applied, the number of edges is counted.

That is, let the number of edges C0UNT in the initial state be O, and b9
), I = 2 (blo) and J = 2 (bll
), the block f“ (I
, J), the absolute value ABS (f“ (1, J)) is the threshold T
It is necessary to judge whether it is larger than H (bl2), and if it is, the number of edges is set as C0UNT=COUNT+1 (bl2).
3) Count up.

Then, J=J+1 and bl4), determine whether J > 7 or not bl5), I=-I+1 and bl6)
, I > 7 (bl7) If J > 7 and I > 7, the counting of the number of edges in the block is completed, so it is determined whether C0UTN<THa (bl7).
8). Then, the number of edges C0UNT is the third threshold TH
If it is 3H3 or less, the number of edges is small, so it is determined that it is a flat portion. Next, it is determined whether TH3≦C0UNT≦TH4 (bl9), and the number of edges C0UNT is set to the third threshold TH.
3 and below the fourth threshold TH4, the number of edges is relatively large, so it is determined that it is an edge part, and the number of edges C0
If UNT exceeds the fourth threshold TH4, the number of edges is greater than when determining it as an edge portion, so it is determined that the pattern is a fine pattern portion.

9 and 10 show flowcharts of image judgment based on the average value, 8×8 block B (8°8), threshold value T
Ha, the fifth and sixth threshold values TH5°TH6, the count value C0UNT and the average value H are determined (C1), and steps (a2) to (alo) similar to steps (alo) shown in FIG. 5 are performed.
The average value H of the 8×8 pixel block is determined by C2) to (clo). And step (cll) ~, (
C19), the pixel value B (1, J,) and the average value H
If the absolute value of the difference between
, J)-Hl>THa) (step (C14)),
Let COUNT=COUNT+1 (C15).

Next, it is determined whether C0UNT<TH5 (C20), and if it is less than or equal to the fifth threshold TH5H5, this means that the value of each pixel in the block is close to the average value, so it is determined that the block is a flat portion. Next, determine whether TH5≦C0UNT≦TH6 (C21)
, if the count value C0UNT is larger than the fifth threshold TH5 and less than the sixth threshold TH6, this means that the number of pixels that are different from the average value H is relatively large, so it is determined that it is an edge portion, If it is larger than the sixth threshold TH6, the number of pixels having a difference with respect to the average value H is greater than when determining an edge portion, and therefore it is determined to be a fine pattern portion.

Figures 11, 12, and 13 show a flowchart of image judgment based on the average value of the small process.
A block of pixels is subdivided, for example, an average value is obtained for a small block (subblock) of 2 × 2 pixels, and the difference between this average value and the pixel value exceeds a threshold value is counted up. Block B (8,
8), threshold THb, 5th. sixth threshold TH5, TH6,
Determine the count value C0UNT and the average value H of 2×2 pixels (
dl), the large block is I, J, and the four pixel positions of the small block are II, JJ (II=0.1. JJ=0.1
), the average value H within the small program is determined by steps (d2) to (dl5), and then steps (dl6) to
(d30), when the absolute value of the difference between the average value H and the pixel value is greater than the threshold THb, C0UNT=COUNT
Count up as shown by T + 1 (d22).

Next, it is determined whether C0UNT<TH5 or not (d31), and if it is less than the fifth threshold TH5H5, it is determined that it is a flat part because there are many pixels with values close to the average value H. Next, TH5≦
It is necessary to determine whether C0UNT≦TH6 (d32), and if it is greater than or equal to the sixth threshold TH6, it means that there are many pixels with a large difference from the average value H, so it is determined that the pattern is a fine pattern, and the fifth threshold TH5 is set. If it exceeds the average value H and is less than or equal to the sixth threshold value TH6, it is determined that the pattern is a fine pattern because there are pixels with a large difference from the average value H to the extent that they are within a small block. Fifth in the case of this small block determination. 6th threshold TH
5, TH6 is the 5th TH6 in the case of large block determination. It is also possible to set a value different from the sixth thresholds TH5 and TH6.

Threshold value THTHa in each pixel determination example described above
THb and the first to sixth thresholds TH1 to TH6 can be set in accordance with application conditions such as when the input image signal contains many still images and many moving images. Also the 6th
In the step (dl9) for determining the standard deviation σ in the figure, the division process by 1/64 is omitted, and the first. Second threshold T
Similar actions and effects can be obtained even if HI and TH2 are set to large values.

In addition, in the image judgment based on the standard deviation in FIGS. 5 and 6, if the standard deviation σ can be assumed to be in the range of 0 to 60, for example, the first threshold THI is set to 10, and the second threshold TH is set to 10.
2 can be 30. In addition, in the image judgment by edge detection in FIGS. 7 and 8, the number of edges is in the range of 0 to 64, but for example, the third threshold TH3 is set to 5, and the fourth threshold TH4 is set to 15. Can be set.

Further, in the image determination based on the average value in FIGS. 9 and 10, the fifth threshold TH5 can be set to 5, and the sixth threshold TH6 can be set to 10.

FIG. 14 shows a flowchart of quantization step control, and the quantization unit 12, 32.42 includes a plurality of quantizers A, B,
A plurality of inverse quantizers A are provided in the inverse quantizer 20 on the receiving side or in the inverse quantizer 37 and 47 for interframe coding.
This corresponds to the case where switching control is performed by providing ", B', and C'. First, an 8×8 pixel block F' (8, 8) and quantizers A, B, and C with different quantization steps, Inverse quantizer A".

B' and C' are determined (el), and an image determination signal is input to perform selection (SEL) (C2). That is, it is necessary to judge whether the judgment signal indicates a flat portion (C3), and in the case of a flat portion, quantizer A with a small quantization step is selected, and similarly, on the receiving side or at the front stage of the frame memory, quantizer A is selected. , selects an inverse quantizer A" corresponding to this quantizer A.

Also, if it is not a flat part, it is difficult to judge whether it is an edge part or not.C
4) In the case of the edge part, the quantization step is performed by quantizer A.
A quantizer B with a quantization step equal to or larger than is selected, and a corresponding inverse quantizer B is selected on the receiving side or at the front stage of the frame memory.Also, if it is not an edge part, Select a quantizer C whose quantization step is the same as or larger than that of quantizer B, and select a corresponding inverse quantizer C'' on the receiving side or at the front stage of the frame memory. .

Instead of making such a selection of quantizers A, B, and C,
It is also possible to adopt a configuration in which the quantization steps in the quantization units 12, 32.42 are switched and controlled, and similarly, the inverse quantization steps can also be switched in the inverse quantization units 20, 37.47. It can be configured to control.

A block determined to be a flat portion is a block in which pixel values change little, and the high frequency components of the transform coefficients obtained by the discrete cosine transform are further reduced, so that even if the high frequency components are deleted, there is less deterioration in the reproduced image quality. Furthermore, since the quantization steps are made finer, the deterioration of the reproduced image quality in flat areas where visual sensitivity is high is reduced. Blocks that are determined to be edge parts, such as the outlines of characters or figures, can be perceived to have some degree of deterioration in the reproduced image quality due to quantization errors in high-frequency components, but the visual sensitivity is lower than that of flat parts, so the quantization step Blocks that are determined to be coarser than those for flat areas, and blocks that are determined to be fine patterns such as striped patterns, will contain relatively many high-frequency components of the transform coefficients obtained by discrete cosine transform, but the visual sensitivity will be reduced. Since this is low, deterioration in reproduced image quality due to quantization errors is almost imperceptible, and therefore the quantization step is made as coarse as or coarser than that for edge portions, thereby improving encoding efficiency.

The present invention is not limited to the above-described embodiments, and the threshold values for image determination are optimally set in consideration of the properties of the image.

〔Effect of the invention〕

As explained above, in the present invention, the image determination section 4
It is determined whether the input image signal corresponds to a block, whether it is a flat part, an edge part, or a fine pattern part, and based on the determination result, the quantization step in the quantization unit 2 is controlled, and the discrete cosine transform unit 1 is applied to the block correspondence. The transform coefficients of the input image signal subjected to the discrete cosine transform are quantized, and the quantized output signal is variable-length encoded in the encoding unit 3, and the quantization step of the transform coefficients is performed according to visual characteristics. The control has the advantage that encoding efficiency can be improved to such an extent that deterioration in reproduced image quality is imperceptible.

Furthermore, even when applying motion compensation to further improve coding efficiency, since the quantization step is controlled by image judgment, it is possible to prevent an increase in block distortion of recurrent images. There are advantages.

The image determining unit 4 determines whether it is a flat part, an edge part, or a fine pattern part based on the size of the standard deviation σ, the number of edges, the number of pixels of the difference from the average value, etc. This can be easily determined during the block-based processing period in the converter 1, and the quantization step of the quantizer 2 can be controlled based on the determination signal to avoid deteriorating the reproduced image quality ( , it is possible to improve encoding efficiency.

[Brief explanation of the drawing]

Figure 1 is a diagram explaining the principle of the present invention, Figures 2, 3, and 4.
Figure 5 is a block diagram of different embodiments of the present invention.
7 and 8 are flowcharts for image determination based on edge detection, FIGS. 9 and 10 are flowcharts for image determination based on average value, and FIG. 11 is a flowchart for image determination based on standard deviation. , FIG. 12 and 1
Figure 3 is a flowchart of image judgment based on small block average values, Figure 14 is a flowchart of quantization step control,
FIG. 15 is a block diagram of the conventional example, FIG. 16 is an explanatory diagram of scanning of blocks, and FIG. 17 is an explanatory diagram of scanning of the conventional example. 1 is a discrete cosine transform section, 2 is a quantization section, 3 is an encoding section, and 4 is an image determination section.

Claims (5)

[Claims] (1) A discrete cosine transform unit (1) that performs a discrete cosine transform on an input image signal or an inter-frame difference signal of the input image signal, and a transform coefficient from the discrete cosine transform unit (1) or the transform coefficient. a quantization unit (2) that quantizes the inter-frame difference signal of the quantization unit (2), an encoding unit (3) that performs variable length encoding of the quantized output signal from the quantization unit (2), and a property of the input image signal. an image determination unit (4) for determining the
The image determining unit (4) determines whether each block of the input image signal corresponds to a flat part, an edge part, or a fine pattern part, and based on the determination result, the quantizing unit (4) 2) An image signal encoding control method characterized by controlling the quantization step. (2) A discrete cosine transform unit (1) that performs a discrete cosine transform on the inter-frame difference signal of the input image signal.
, a quantization unit (2) that quantizes the transform coefficients from the discrete cosine transform unit (1), and an encoding unit (3) that performs variable length encoding on the quantized output signal from the quantization unit (2). ), an image determination unit (4) that determines the properties of the input image signal, and a motion compensation unit that performs motion compensation for blocks between frames of the input image signal, and the image determination unit (4) The image signal encoding control system according to claim 1, characterized in that the quantization step of the quantization section (2) is controlled. (3), the quantization unit (2) by the image determination unit (4);
) is to make the quantization step finer when it is determined to be the flat part, to make it coarser than the quantization step for the flat part when it is determined to be the edge part, and to make the quantization step coarser than the flat part when it is determined to be the edge part. 2. The image signal encoding control method according to claim 1, wherein when it is determined that the quantization step is the same as or coarser than the quantization step in the edge portion. (4) The image determining unit (4) determines that the standard deviation of the pixel values of the block is a flat portion when it is less than or equal to a first threshold, and when the standard deviation of the pixel values of the block exceeds the first threshold, it is determined that the standard deviation of the pixel values of the block is a flat portion. 2. The image signal encoding control system according to claim 1, wherein the image signal encoding control system determines that the image signal is an edge portion when the following conditions occur, and determines that the image signal is a fine pattern portion when the second threshold value is exceeded. (5), the image determination unit (4) applies a Laplacian mask to the pixels of the block, and determines it to be a flat portion when the number of pixels exceeding a predetermined threshold is equal to or less than a third threshold; 2. The image signal code according to claim 1, wherein the image signal code is determined to be an edge portion when it exceeds the third threshold and is equal to or less than a fourth threshold, and is determined to be a fine pattern portion when it exceeds the fourth threshold. control method. (6) The image determination unit (4) calculates the average value of the pixel values of the block or subblocks obtained by subdividing the block, and combines the average value with the value of each pixel in the block or subblock. The number of pixels for which the absolute value of the difference exceeds a predetermined threshold is counted, and when the number of pixels is less than or equal to the fifth threshold, it is determined to be a flat part, and when the number of pixels exceeds the fifth threshold, it is determined to be a flat part. 2. The image signal encoding control method according to claim 1, wherein when the number of pixels is equal to or less than a threshold value, it is determined to be an edge portion, and when the number of pixels exceeds the sixth threshold value, it is determined to be a fine pattern portion.