JPH0595543A - Video coding / decoding device - Google Patents

Abstract

(57) [Summary] [Object] The present invention provides a moving picture coding / decoding apparatus,
An object of the present invention is to propose a moving picture coding / decoding apparatus capable of obtaining a high quality transmission image with a small amount of information. [Structure] Images containing discontinuous image information, such as edges,
By switching to predictive coding, it is possible to perform coding with the same or a smaller amount of information as DCT coding and less interference such as mosquito noise, and a smaller amount of information and a higher total amount of information than a system using only DCT coding. Image quality is obtained. Furthermore, when predictive coding is performed, the block distortion of the decoded image can be reduced by transmitting the predicted value of the block and the quantization width (or the dynamic range of the block) and performing adaptive quantization. In addition, coding efficiency can be further improved by performing DPCM after quantization.

Description

Detailed Description of the Invention

[0001]

[Table of Contents] The present invention will be described in the following order. Field of Industrial Application Conventional Technology Problems to be Solved by the Invention Means for Solving the Problems (FIGS. 1 and 2) Action (FIGS. 15 to 18) Example (FIGS. 1 to 22) (1) Overall Configuration (FIGS. 1 and 2) (2) Configuration of Encoding Device (FIG. 1) (2-1) Configuration of DCT Processing Unit 5 (FIGS. 3 and 4) (2-2) Predictive Encoding Processing Unit 6 Process (2-3-1) Method of using average value of image signal in block as prediction value (FIGS. 5 to 7) (2-3-2) Prediction value is calculated using adaptive dynamic range coding Method (FIGS. 8 and 9) (2-3-3-1) Edge matching quantization method (1)
(FIGS. 10 to 12) (2-3-3-2) Edge matching quantization method (2)
(FIG. 13) (2-4) Configuration of DPCM Circuit 25 (2-4-1) DPCM System (1) (FIG. 14) (2-4-2) DPCM System (2) (FIG. 14) (2-4 -3) DPCM method (3) (FIGS. 15 to 18) (2-5) Processing of DCT / prediction coding determination circuit 7 (2-5-1) Determination in spatial domain (2-5-2) DCT Determination in Transform Output Area (FIG. 19) (3) Configuration of Decoding Device (FIG. 2) (3-1) Configuration of Inverse DPCM Circuit 27 (FIGS. 20 to 2)
2) (4) Operation of the embodiment (5) Effect of the embodiment (6) Other embodiment Effect of the invention

[0002]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is suitable for application to a moving picture coding apparatus that codes a moving picture with high efficiency and transmits the moving picture.

[0003]

2. Description of the Related Art Conventionally, in a video signal transmission system for transmitting a moving image to a remote place such as a video conference system or a video telephone system, in order to efficiently use a transmission line, the video signal is transmitted between frames or within a frame. It is designed to enhance the transmission efficiency of important information by encoding. A typical two-dimensional Discrete Cosine Transform (DCT) system is a typical coding system for highly efficient coding of such coded data.

This Discrete Cosine Transform method is
The amount of information is compressed by concentrating signal power on a specific frequency component using the two-dimensional correlation of an image signal and encoding only the concentrated distribution coefficient. For example, the distribution of DCT coefficients concentrates on low-frequency components in a portion where the pattern is flat and the autocorrelation of the image signal is high. In this case, the amount of information is compressed by encoding only the coefficients distributed in the low frequency range. Is made possible.

On the other hand, in a pattern with many edges, the coefficients are widely dispersed from low frequency components to high frequency components at the discontinuity points of the edges. In this case, in order to accurately represent the discontinuous point of the image signal with the DCT coefficient like an edge, a very large number of coefficients are required, so that the coding efficiency is lowered. At this time, if the quantization characteristic of the coefficient is roughened for high compression coding of the image or the coefficient of the high frequency component is discarded, the deterioration of the image signal becomes noticeable. For example,
The ringing becomes noticeable visually.

On the other hand, there is a predictive coding method as a moving picture coding method. With this predictive coding method, a transmission image can be coded with relatively high efficiency for a correlated pattern. However, when the correlation is small, there is a problem that the amount of information increases and the transmission efficiency decreases. Therefore, it has been proposed to improve the transmission efficiency by switching the encoding method of the transmission image in block units.

[0007]

However, when trying to predictably code a transmission image in block units, there is a problem that block distortion is conspicuous. Here, the block distortion means a phenomenon that an image partially looks like a mosaic for each block as a result of encoding.

This is because the predictive coding compresses the redundancy of only the amplitude of the image signal, and block distortion is likely to occur when the compression rate is increased as compared with the discrete cosine transform method which compresses up to the spatial frequency domain. This is because. In this way, when switching between the discrete cosine transform method and the predictive coding method in block units,
When the image is encoded and transmitted by simply switching to the one having higher encoding efficiency, the encoding efficiency is improved, but there is a problem that the image is deteriorated due to block distortion of the decoded image.

The present invention has been made in consideration of the above points, and an object thereof is to propose a moving picture coding apparatus capable of obtaining a high quality transmission image with a small amount of information.

[0010]

In order to solve such a problem, in the first invention, an image is divided into unit blocks corresponding to a plurality of pixels, and the pixel data of the pixels is orthogonally transformed by each unit block. In the moving picture coding / decoding apparatus 1 for decoding and transmitting, the first picture element in the unit block is initially set to the first picture element when the picture element data is transmitted by predictive coding. From the position to a second position separated by a predetermined pixel,
By obtaining the difference value between the second pixel data and the pixel data of the first pixel around the position of, the image information of the image is compressed and transmitted.

According to the second aspect of the present invention, a moving image coding is performed in which an image is divided into unit blocks corresponding to a plurality of pixels, and pixel data of the pixels is orthogonal transform coded or predictively coded for each unit block and transmitted. In the moving picture coding / decoding apparatus 1 for decoding and transmitting, when transmitting pixel data by predictive coding, the pixels in the unit block are moved by a predetermined pixel in the same row or in the same column in row units or column units. The second row or column adjacent to the second position after the movement.
By calculating the difference value between the pixel data of the pixel and the pixel data of the pixel, the image information of the image is compressed and transmitted.

[0012]

When one image is divided into unit blocks and each unit block is coded, the discrete cosine transform coding and the predictive coding are switched according to the nature of the pattern, and the predictive coding in the block unit. In order to reduce the block distortion (discontinuity between blocks) caused by, the coding unit predicts and predicts the unit block prediction value and quantization width, or the unit block prediction value and the dynamic range. , And adaptively quantize, and in the decoding device, switching information, block prediction value and quantization width,
Alternatively, a high-quality image can be obtained by decoding the image based on the block prediction value and the dynamic range.

[0013]

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

(1) Overall Structure In FIGS. 1 and 2, reference numeral 1 denotes an image data transmission system as a whole, and an encoding device 2A encodes image data within a frame (field) or between frames (field) and after encoding. The transmission image is transmitted by appropriately switching the discrete cosine transform coding process or the predictive coding process for each block.

The coding device 2A adds this switching information to the switching information flag for discrete cosine transform coding or predictive coding for each area to be coded, or an information flag indicating the case classification of the area to be coded. Is extended to provide a mode of predictive coding. Further, the decoding device 2B is adapted to decode the transmission image by using these information flags transmitted together with the encoded image data.

(2) Configuration of Encoding Device The encoding device 2A receives the image data of 8 × 8 pixels as the input digital data S1 and performs the discrete cosine transform via the difference data generating circuit 3 and / or the switching circuit 4. The processing unit 5, the predictive coding processing unit 6, and the DCT / predictive coding determination circuit 7 are supplied to the respective units. The difference data generation circuit 3 uses the input digital data S1.
And the difference between the prediction data S2 input from the prediction circuit 8 and the difference data S3 are output to the switching circuit 4.

The switching circuit 4 uses the inter-frame / intra-frame coding switching signal S4 input from the prediction circuit 8 to input the input digital data S1 when the intra-frame coded data can be transmitted with a smaller data amount. The difference data S3 is output when it can be output as it is, or when it can be transmitted with a smaller data amount by inter-frame encoding and transmission.

Here, the DCT processing section 5 utilizes the two-dimensional correlation of the input image to perform the discrete cosine transform of the input image data S1 or the difference data S3 in the unit of a minute block, and the transformed data obtained as a result is subjected to a predetermined conversion. When quantized with the quantization size, the variable length coding circuit 1 is passed through the switching circuit 9.
It is designed to output to 0.

Further, when the predictive coding unit 6 predictively codes the image signal in the block, the predictive coding unit 6 obtains the difference between the predicted value and the current image signal, quantizes the difference signal with a predetermined quantization size, and switches it. It is adapted to output to the variable length coding circuit 10 via the circuit 9. The DCT / predictive coding determination circuit 7 determines whether to perform discrete cosine conversion in block units or predictive coding in block units when coding an image signal, and the determination result is switched to DCT / predictive coding. The signal S5 is output.

As a result, the switching circuit 9 outputs the conversion data S6 output from the DCT processing unit 5 when the discrete cosine conversion data is to be output.
When the predicted coded data is to be output to 0, the converted data S7 output from the predictive coding processing unit 6 is output to the variable length coding circuit 10.

Here, the variable length coding circuit 10 has a conversion table corresponding to the DCT conversion data S6 and the predictive coding conversion data S7 having different statistical properties and switching controlled by the DCT / prediction coding switching signal S5. Then, the conversion efficiency is further improved by the conversion table and output as the transmission data S8.

The buffer circuit 11 also transmits the transmission data S8.
After it is stored in the memory, it is output in a predetermined order, and a quantization width control signal for controlling the quantization size is output so that the residual data remaining in the memory has an appropriate remaining amount. ing. Here, the encoding device 2A is
D through the inverse quantization circuit 12 and the inverse DCT circuit 13 in sequence
The quantized data S10 input from the CT processing unit 5 is inversely quantized into a representative value, and the locally decoded image data S11 is further subjected to the inverse transform process of the discrete cosine transform.
It is designed to be converted to and output.

Further, the encoding device 2A includes an inverse quantization circuit 14
Then, the quantized data S12 input from the predictive encoding processing unit 6 through the inverse predictive encoding circuit 15 is inversely quantized into a representative value, and further the decoded image data S13 is obtained by the conversion processing opposite to the predictive encoding processing. It is designed to be converted to and output. The switching circuit 16 outputs the decoded image data S11 when the input image is DCT-converted and transmitted based on the DCT / predictive coding switching signal S5, and the input image is predictively coded and transmitted. In such a case, the decoded image data S13 is converted and output to the addition circuit 17.

Here, the adder circuit 17 adds the image data S14 input from the switching circuit 18 and the decoded image data S11 or S13 to obtain the locally decoded data S15 as the prediction circuit 8.
It is designed to be supplied to. Here, the prediction circuit 8
The prediction data S2 is supplied to the difference data generation circuit 3 and the switching circuit 18 based on the locally decoded data S15, and
The motion vector / prediction mode determination signal S16 is output to the variable length coding circuit 10 based on the locally decoded data S15.

(2-1) Configuration of DCT Processing Unit 5 The DCT processing unit 5 has a DCT circuit 20, a quantization circuit 21 and a delay circuit 22, and the DCT circuit 20 uses the two-dimensional correlation of the input image. And input digital data S1
Alternatively, the difference data S3 is converted into the discrete cosine coefficient data S21 and output to the quantization circuit 21 (Journal of the Institute of Electronics, Information and Communication Engineers 1987 / 1Vol.J70-
B No. 1 p96-104 “Intra-frame / inter-frame adaptive extrapolation interpolative coding of HDTV signal” and Japanese Patent Application No. 2-41024
No. 7).

The quantizing circuit 21 quantizes the discrete cosine coefficient data S21 in accordance with the amount of generated information and outputs the quantized data S10 to the delay circuit 22 and the inverse quantizing circuit 12. Delay circuit 22
Is configured to delay the quantized data S10 by a delay time corresponding to the processing time of the predictive coding processing unit 6 and output the quantized data S10 to the variable length coding circuit 10.

Incidentally, it is known that the DCT method generally has a property that a large value is concentrated around a certain coefficient by performing DCT on a signal whose brightness changes smoothly. For example, as shown in FIG. 3 (A), an original image having a luminance level of 30 to 100 in which each pixel is a pattern including edges from the upper left to the lower right as one block of 8 × 8 pixels is input. In this case, when the discrete cosine transform is performed by the DCT circuit 40, most of the coefficients in the block become 0 (FIG. 3 (B)). Further, the coefficient having the value exists on the diagonal line from the upper left to the lower right.

Next, when this coefficient is quantized by the quantizer circuit 21 with a brightness level of 10, for example, most of the coefficients become 0 and only large coefficients remain. Therefore, the coefficients are sequentially called according to the calling sequence of the coefficients shown in the figure (the numbers indicate the calling order), and the variable length coding circuit (VL
High efficiency coding can be performed by applying a variable length coding method such as Huffman code according to C) 10.

At this time, taking into consideration that pixels have a correlation in a two-dimensional direction, it has been proposed to call a coefficient that is gradually distant from a coefficient calling start point. Further, in the case of a pattern having a strong coefficient correlation in the horizontal direction, the coefficients in the upper stage (0 to 7) of the block can be sequentially called in the reading order as shown in FIG. 4, and subsequently can be called as (8 to 63).

(2-2) Processing of Predictive Encoding Processing Unit 6 The predictive encoding processing unit 6 has a predictive encoding circuit 23, a quantizing circuit 24 and a DPCM (Differntial PCM) circuit 25. When the predictive coding processing unit 6 predictively codes the image signal in the block, the predictive coding processing unit 6 generates a differential signal of the block from the image signal information in the block and quantizes it to generate DPCM.
High-efficiency coding is performed by reducing autocorrelation left in the differential signal by the circuit 25 and coding the data by the variable-length coding circuit 10 (JP-A-1-135281). JP-A-2-134910
Publication).

At this time, the buffer circuit 11 monitors the output data and, depending on the amount of information, the quantization width Q of the predictive coding circuit 23.
Is made to decide. Specifically, the following three methods can be used as the adaptive quantization method. In the present invention, one or more of these methods are selectively used for each block.

(2-3-1) Method of using the average value of the image signal in the block as the predicted value, that is, as shown in FIG. 5, the average value of the amplitudes of all the pixels in the block is obtained, and the average value and the average value are respectively obtained. This is a method of quantizing the difference in signal level between pixels. Here, the quantization width Q is a value output from the buffer circuit 11. At this time, if the average value in the block is M and the signal level of the pixel in the block is L, the quantization code Lq is

[Equation 1] It is represented by. Here, // indicates rounding off to the first decimal place.

If the restoration value is L ',

[Equation 2] Can be expressed as (FIG. 6). However, in this method, when the quantization width Q is large, the restoration value distortion becomes large, and discontinuity between blocks tends to occur as shown in FIG.

(2-3-2) Method of Obtaining Prediction Value Using Adaptive Dynamic Range Coding (ADRC) This method is based on the "quantization method of adaptive dynamic range coding". Kondo et al., 1989, 4th Image Coding Symposium (PCSJ) Material (4-3).

Adaptive Dynamic Range Coding ADR
FIG. 8 shows the quantization characteristic when C is applied, and FIG. 9 shows an example of the block distortion caused by the adaptive dynamic range coding ADRC. In the adaptive dynamic range coding ADRC, the reason why the minimum value in a block is used as a prediction value is that the minimum value is often located in the peripheral portion of the block.

That is, since the normal block is a small area of about 8 × 8 pixels, when the brightness level is concave, it is extremely low. For this reason, the minimum value of a certain block is often close to the minimum value of any of the surrounding blocks. Therefore, when the minimum value is in the peripheral portion of the block, the continuity with at least one peripheral block is maintained, and the block distortion is suppressed to the minimum.

Further, in the adaptive dynamic range coding ADRC, as shown in FIG. 8, the average value of the signal values included in the highest and lowest gradation levels is newly added to the maximum value.
Redefining the minimum value makes it less susceptible to noise and isolated points (JP-A-2-134910).

(2-3-3-1) Edge Matching Quantization Method (1) First, the one-dimensional case will be described. With respect to the block length signal as shown in FIGS. 10 and 11, the signal values at both ends of the block are set to X1 and X2, and X1 ≦ X2 is set for simplification, and the restoration values of the signal values X1 and X2 are determined. The quantization width Q output from the buffer 11 is changed by the following algorithm so that it is output below the error.

If the signal level of the pixel in the block is L and the restoration error tolerance of the signal values X1 and X2 is Ex, the dynamic range DR is

[Equation 3] And the quantization width Q and the signal value X1 are

[Equation 4] And,

[Equation 5] Then, the quantization width Q is set to the same value as the quantization width q indicated by the quantization width control signal S22 output from the predictive coding circuit 23 to the quantization circuit 24, and the signal value X1 is left unchanged.

[0040] Also

[Equation 6] And

[Equation 7] Then, the quantization width Q and the signal value X1 are left unchanged. Here, ABS represents an absolute value.

Where

[Equation 8] Then, the quantization width Q is left unchanged and the signal values X1 and X
When the average value of 2 is Xm = (X1 + X2) // 2, the signal value X1 is Xm. However, the quantization width q specified by the predictive coding circuit 23 is defined as follows for all the quantization width Q and dynamic range DR.

[Equation 9] The table is written in the ROM by finding a value larger than Q that satisfies the above condition.

At this time, the quantization code Lq is

[Equation 10] And the restored value L'is

[Equation 11] Given by. In this method, since the error of the restored signal at both ends of the block is suppressed by Ex, the continuity between blocks can be further maintained.

Next, a two-dimensional case will be described. To extend the one-dimensional method to a two-dimensional block, X1, X
2 A sub-routine for making a decision is further required. The algorithm for this determination is as follows. First 8x
As for the two-dimensional block signal of 8 pixels, as shown in FIG. 12, areas 1a, 1b, 2a, 2b, 3a, 3b, 4 at the end of the block
Consider a and 4b.

Next, the average value of the pixel values in each area is calculated for each area. Here, each of these average values is m
1a, m1b, m2a, m2b, m3a, m3b, m4
a and m4b. After this, ABS (m1a-m1
b), ABS (m2a-m2b), ABS (m3a-m)
3b) and ABS (m4a-m4b), the largest one is obtained.

By the above procedure, the direction in which the pixel values greatly differ at both ends of the block, that is, the direction across the edge is obtained. Therefore, the smaller one of the representative values of the pixels in the selected two regions is set to X1, the other is set to X2, and the block signal is quantized in the same manner as in the one-dimensional case.
Dequantize.

(2-3-3-2) Edge Matching Quantization Method (2) First, the one-dimensional case will be described. For a block length signal as shown in Fig. 13, set the signal values at both ends of the block to X.
1 and X2. For simplicity, X1 ≦ X2. This method is characterized in that the decoder side changes the restoration values so that the signal values X1 and X2 are output as the restoration values as they are. When the signal level of the pixel in the block is L, and the quantization width output from the buffer 11 is Q, the quantization code Lq is

[Equation 12] Becomes

Here, in the encoding device 2A, the quantized code L
In addition to q, the signal values X1, X2 and the quantization width Q are transmitted as quantization parameters. In the decoding device 2B, the signal values X1 and X2 and the quantization width Q are received as the quantization parameter, and first, the quantization value X of the signal value X2 is received.
It is designed to calculate 2q.

[Equation 13]

After this, the quantization code Lq becomes the quantization value X2.
If it is equal to q,

[Equation 14] And if they are not equal,

[Equation 15] Is restored to the restored value L ′. This method has a simpler algorithm than the edge matching quantization method (1) described above, and does not require a ROM table for changing the quantization width Q.

Next, a two-dimensional case will be described. When the edge matching quantization method (2) is extended to a two-dimensional block signal, X1 and X2 are determined by the method described in the edge matching quantization method (1). Subsequent quantization and decoding methods are performed in the same manner as in the one-dimensional case.

(2-4) DPCM (Differntial PCM)
Processing of Circuit 25 The coefficients after adaptive quantization still have almost the same autocorrelation of the signal that the original image signal had. Therefore, the amount of information can be further compressed by performing DPCM in the subsequent stage. In the DPCM circuit 25, the scan direction of the input pixel is designated by the scan signal S24 from the predictive coding circuit 23.

The DPCM circuit 25 obtains the difference between the quantized image signal Yi input by the differentiator and the pixel value one pixel before,

[Equation 16] Is designed to get you.

The prediction error signal Ei can take values from -255 to +255 when the input signal is 8 bits. Therefore, if it is attempted to transmit the data as it is, 9 bits are required, and an extra code is required for each pixel. However, it is known that the prediction error signal is mostly centered around zero and before and after it. Therefore, not all signals are represented by 9 bits, but by assigning a code with a short bit length to the signal values that appear in large numbers, the block as a whole will have an average of 9 bits as well as the original input signal of 8 bits. A block signal can be represented with a bit length much shorter than that.

Here, the following three methods can be used as a method suitable for the DPCM in the block. In the present invention,
One of the following methods is selectively used for each block.

(2-4-1) DPCM method (1) DPCM of the block signal after adaptive quantization is performed in the calling path as shown in FIG. Since there is only one call path, the circuit configuration is simple. However, except for the block including the contour like the mode 1 in FIG. 14A (mode 2 in FIG. 14B, mode 3-1 in FIG. 14C, mode 3-2 in FIG. 14D). Therefore, it is not possible to expect a significant amount of information compression.

(2-4-2) DPCM method (2) Then, it is considered to switch the data calling method for each of the modes 1 to 3-2 so that the optimum amount of information can be compressed. 14A to 14D show calling methods suitable for each of the modes 1 to 3-2. Next, a method of determining mode switching of these DPCM will be described. This determination is made by supplying the DPCM circuit 25 with the information obtained by the examination of the directionality of the contour in the block signal described in the edge matching quantization method (1).

That is, when i = 1, mode 1
= 2, mode 2; i = 3, mode 3-1
, And when i = 4, it corresponds to mode 3-2.

(2-4-3) DPCM System (3) This system utilizes the feature that most of the objects of DPCM are blocks containing edges. The presence of an edge on a block means that discontinuities are arranged in the block in any of vertical, horizontal, or diagonal directions. Of course, the arrangement is not limited to a straight line, and the lines are arranged, for example, as shown in FIG.

Even if such a block is encoded by the DPCM method (1) or the DPCM method (2), the feature of the edge does not coincide with the scan direction, so that the difference value becomes large and the encoding efficiency is low. To solve this problem, the difference is calculated by the following algorithm. For concrete description, the block size is 8 × 8 pixels, and DPCM is performed in the vertical direction. (Actually, the block size can be any number, and the DPCM direction can be horizontal.)

That is, for the first row, DPCM is performed in the horizontal direction for each pixel to obtain the difference. For the subsequent second and subsequent rows, DPCM is performed in the vertical direction for 8 pixels in the horizontal direction, that is, collectively for each row.

At this time, as shown in FIG. 15B, if the difference is taken in the vertical direction as it is, the difference value becomes large between the third pixel and the sixth pixel. To avoid this, if a difference is calculated after shifting the first row to the right by one pixel as shown in FIG. 15C, a large difference does not occur.

At this time, the problem is the pixels at both ends of the block. However, as shown in FIG. 15D, when the pixel is shifted to the left by 2 pixels, the first and second pixels are not used for prediction.
Instead, the 8th pixel is used to obtain the difference from the 6th, 7th, and 8th pixels in the next row. This is a so-called arithmetic shift.

Since the shift amount is variable for each row as described above, this system will be referred to as a variable shift DPCM. If this variable shift DPCM is used, FIG.
Even blocks containing more complicated edges can be encoded with high efficiency. Furthermore, the shift amount of this variable shift DPCM is not limited to the integer shift amount in pixel units, and the shift amount below the decimal point can be considered. The shift below the decimal point corresponds to interpolating the value between pixels using an interpolation filter. For example, a shift of 0.5 pixel means finding a value at an intermediate point between pixels.

Here, the DPCM circuit 45 has a variable shift D
When encoding using PCM, it is necessary to transmit the shift amount of each row (column) in addition to the difference value. For example, if the shift amount per row (column) is ± 3 pixels, most of the edges can be dealt with, so the shift amount is -3 to +3.
If there are seven types of pixel ranges, 3 bits are required. In that case, for a block of 8 × 8 pixels,

[Equation 17] As shown in, the additional information of 21 bits is required.

Further, by obtaining the probability of occurrence of the shift amount used in the actual image, the amount of additional information can be reduced by the variable length coding. For example, when the probability of occurrence of a shift amount of ± 2 pixels or more is small, by using the symbols in FIG.

[Equation 18] Then, the amount of additional information can be reduced to near 14 bits.

The shift amounts from the first row to the seventh row are correlated with each other. For example, when the edge is a straight line such as a vertical line, a horizontal line, or a diagonal line, the shift amount from the first line to the seventh line has the same value. Therefore, there is a method of arranging the shift amounts from the first row to the seventh row and collectively encoding them.

In this case, particularly when the probability of occurrence of the shift amount 0 is large, if a row of the shift amount 0 continues, it is regarded as a 0 run and the code including the 0 run as shown in FIG. 17 is used. Thus, it is possible to reduce the amount of additional information.
Such difference information and shift amount can be multiplexed and transmitted as shown in FIG.

(2-5) Process of DCT / Predictive Coding Determination Circuit 7 When the image signal is coded, the DCT / predictive coding determination circuit 7 performs DCT in block units or predictive coding. It is determined whether or not to convert. First, it is necessary to judge which processing method should be selected based on the block information. In that case, the determination is made in the spatial domain or the DCT transform output domain. Each determination method will be described in detail below.

(2-5-1) Judgment in Spatial Area For a picture whose brightness changes abruptly, specifically, an image including edges and details, the dynamic range (maximum value-minimum value) of the image signal in the block is determined. Value) takes a large value.
For such a pattern, the DCT has a disadvantage in the compression rate of information, and the predictive coding should be selected. Therefore, for each block, the dynamic range (DR) within that block is obtained, and it is determined that the predictive coding is performed for the block whose value exceeds an appropriate threshold value (A) selected from the compression rate and the deterioration of the pattern. To do.

(2-5-2) Judgment in DCT Transform Output Area It is known that the DCT coefficient when the image signal is two-dimensionally DCT has the following properties. For example, in a two-dimensional DCT having (8 × 8) pixels as a block (macro block), the coefficient F at the 0th row and 0th column corresponding to the upper left corner of the block
(0, 0) corresponds to a DC component that represents the average brightness in the image block. The coefficient represents the vertical high frequency component in the image block as it goes from F (0,0) to the right and horizontal direction, and the horizontal high frequency component as it goes downward.

That is, when a block of a pattern whose brightness changes abruptly like an edge is transformed by discrete cosine conversion, the converted output is largely as shown in FIG.
Can be classified into two. FIG. 19 shows the output area after the discrete cosine conversion in the macro block of 8 × 8 pixels. Here, ◯ indicates the position of a pixel with high (or low) brightness, and × indicates the position where a large DCT coefficient is likely to occur in the block.

FIG. 19A shows the case where the contour exists in the vertical direction, and the DCT coefficients are concentrated with a large energy in the horizontal direction from the low order. This is hereinafter referred to as "case 1". Further, in FIG. 19B, when the contour exists in the horizontal direction, the DCT coefficient concentrates with a large energy in the vertical direction from the low order. This is hereinafter referred to as "case 2".

FIG. 19C shows the case where the contour exists in the oblique direction, and the DCT coefficient concentrates with a large energy in the oblique direction from the low order. Hereinafter, this will be referred to as "case 3". All DC except DC components
Absolute value sum Fa of T coefficient and D of the DCT output area in each of the cases 1, 2, and 3
The absolute value sums F1, F2, and F3 of CT coefficients are obtained for each block.

When the maximum of the absolute value sums F1, F2, and F3 is set to Fmax, a block in which the ratio of Fmax to Fa exceeds an appropriate threshold value selected from the compression rate and the deterioration of the pattern is predictively coded. It is adapted to be encoded by the method.

(3) Configuration of Decoding Device On the other hand, the decoding device 2B, as shown in FIG. 2, receives the encoded bit stream input S3 output from the encoding device 2A.
1 is input to the buffer 31 and is temporarily stored. The inverse variable-length coding circuit 32 decodes from the switching signal whether the transmission data is DCT conversion data or predictive coding conversion data from the coding bit stream input S31, and according to the information, DCT or predictive coding. Or select.

Here, the inverse variable length encoding circuit 32 is adapted to inverse variable length encode the encoded bit stream input S31 and supply it to the inverse DCT processor 33 and the inverse predictive encoding processor 34. .. The inverse DCT processing unit 33 has a delay circuit 35, an inverse quantization circuit 36, and an inverse DCT circuit 37, and decodes a transmission image from the encoded bit stream input S31 by a processing procedure reverse to that of the DCT processing unit 5. ing.

Further, the inverse predictive coding processing section 34 uses the inverse DPC.
M circuit 38, inverse quantization circuit 39, inverse prediction encoding circuit 40
And the transmission image is decoded from the encoded bit stream input S31 by a processing procedure reverse to that of the predictive encoding processing unit 6. Here, the inverse quantization circuits 36 and 3
Reference numeral 9 denotes the quantization width control signal S from the inverse variable length coding circuit 32.
32 and S33 are input.

The inverse DPCM circuit 38 is adapted to receive the scan direction specifying signal S34 from the inverse variable length encoding circuit 32, and the inverse predictive encoding circuit 40 is adapted to input the predicted value signal S35. ing. The decoding device 2B has the decoded data S36 decoded by the inverse DCT processing unit 33 and the inverse predictive coding processing unit 34 by the processing procedure.
And S37, the transmission image is decoded through the switching circuit 41 and the adding circuit 42 in order.

Further, at this time, the predicting circuit 43 is for switching control by the inverse variable length encoding circuit 32, and for reproducing the moving picture from the output of the inverse DCT or inverse predictive encoding processed for each block. Is. Here, the output of the prediction circuit 43 is added via the switching circuit 44 to the addition circuit 42.
Is input to the switching circuit 41,
42 are switched by the DCT / prediction coding switching signal S39 input from the inverse variable length coding circuit 32.

(3-1) Structure of Inverse DPCM Circuit 27 The transmitted encoded data encoded by the DPCM method (3) is decoded by the variable shift DPCM decoder circuit 50 shown in FIG. Has been done. That is, the variable shift DPCM decoder circuit 50 has 8 bits × 1.
It is constituted by a shift register 51 of one stage and three sets of flip-flops 52, 53, 54 constituting a counter.

Here, the variable shift DPCM decoder circuit 5
0 inputs the pixel data S51 (FIG. 21 (B)) raster-scanned in block units to the adder 56 via the D-flip flop (FIG. 21 (F)).

The adder 56 adds the output S56 (FIG. 21 (G)) output from the 8-byte-1 byte selector 58, and outputs the addition result via the AND circuit 57 to 8 bits.
While supplying to the x11-stage shift register 51,
It is adapted to be output as raster scan pixel data output via the flip-flop circuit 59.

Here, the raster scan pixel data output from the D-flip-flop circuit 59 is delayed by two clocks from the input clock signal S50 (FIG. 21A) for one pixel. Is output. Further, as shown in FIG. 22, the shift registers 52, 53 and 54 have 3-bit shift amount designation inputs S00, S01 and S02 (FIG. 21) for designating seven types of shift amounts.
(E)) is inputted via the logic operation circuits 60, 61 and 62, respectively.

Here, each shift register 52, 53 and 5
4 is operated by the clock signal S50 and the input terminal LD
A flag S53 (FIG. 21 (D)) indicating the beginning of each line is input to. Also shift register 5
A flag S52 (FIG. 21C) indicating the beginning of each block is input to the clear input terminals CLR of 2 and 53.

At this time, the shift registers 52, 53 and 54 are respectively arranged in FIG. 21 (H), FIG. 21 (K) and FIG.
1 (L), the logical product of the outputs (FIGS. 21 (I) and (J)) from the RCO output terminals of the shift registers 53 and 54 is input to the ENP input terminal of the counter 52. As a result, the outputs QA, Q of the shift register 52
Switching signal S57 of 3 bits from B, QC to selector 58
(FIG. 21 (H)) is supplied.

(4) Operation of the Embodiment In the above configuration, the encoding device 2A switches the switching circuit 4 by the inter-frame / intra-frame switching signal S4 input from the prediction circuit 8 so that a frame is transmitted in accordance with the transmitted image. The inter-frame coded or intra-frame coded image data is output to the DCT processing unit 5 and the predictive coding processing unit 6.

At this time, the DCT processing unit 5 performs discrete cosine transform on the image data of 8 × 8 pixels by using the two-dimensional correlation, and the transformed image data is further controlled by the quantization width control input from the buffer circuit 11. Quantize based on the signal S9. On the other hand, the predictive coding processing unit 6 quantizes the predictive-coded image data through the predictive coding circuit 23 and the quantizing circuit 24 in a predetermined quantization width to quantize the quantized data S10 into the DPCM circuit. Supply to 25.

Here, the DPCM circuit 25 encodes each block according to each of the DPCM system (1), the DPCM system (2), and the DPCM system (3), together with a flag indicating which conversion system has been encoded. The shift amount and / or difference data at that time is output to the variable length coding circuit 10 in the subsequent stage. Here, encoding results are compared for each encoding method for blocks having a dynamic range of 50/255 or more, which is given by 8 × 8 pixels as a transmission image.

First, the target block is DPCM-encoded in the vertical and horizontal directions, and the one having the smaller sum of absolute values of the remaining amount after encoding is set as the encoding target amount of the one-way DPCM encoding.
Subsequently, similarly, variable shift DPCM coding by the DPCM method (3) is similarly performed on the same block, and the coding target amount at that time is obtained.

At this time, if the blocks whose coding target amount by the variable shift DPCM coding is less than 60% of the coding target amount by the unidirectional DPCM coding are counted,
Blocks for which variable shift DPCM coding is advantageous for various transmission images occupy 6% to 35%. In particular, when the transmission image includes many edges in the oblique direction, the variable shift DPCM coding method is used to perform the coding, and the coding is 40% compared to the case where the one-way DPCM coding method is used. The above data amount can be reduced.

As a result, the DPCM circuit 23 performs variable length coding on the image data coded by the variable shift DPCM coding method according to the DPCM method (3) for the blocks in which the diagonal edge is included in the same block. Circuit 1
Output to 0.

(5) Effects of the Embodiments According to the above configuration, by switching to predictive coding at a signal discontinuity point such as an edge, a mosquito noise or the like can be generated with the same or a small amount of information as DCT coding. It is possible to perform coding with less interference, and it is possible to transmit high image quality with a smaller amount of information as a whole as compared with a system only using DCT coding. In addition, when predictive coding the transmitted image and transmitting it, the block prediction value and the quantization width (or the dynamic range of the block) are transmitted, and adaptive quantization is performed to reduce the block distortion of the decoded image. You can

Furthermore, by performing DPCM after quantization, the coding efficiency can be further improved. DPCM
Since the coding method according to the method (3) is completed on a block-by-block basis, it can be realized with a smaller hardware scale than the conventional predictive coding method.

(6) Other Embodiments In the above embodiment, when the difference data is obtained by the DPCM method (3), the data in each row is shifted to the left and right, and the shift amount and the difference data at that time are transmitted. Although the case has been described, the present invention is not limited to this, and the data shift amount and the difference data may be obtained for each column and transmitted.

In the above embodiment, the case where the present invention is applied to the encoding devices 2A and 2B shown in FIGS. 1 and 2 has been described, but the present invention is not limited to this, and predictive encoding of image data is performed. The present invention can be widely applied to an encoding / decoding device that transmits by transmitting.

[0095]

As described above, according to the present invention, the coding method is switched to the discrete cosine transform coding or the predictive coding for each unit block, and the transmission image is coded.
In the case of predictively-coded image transmission, when the difference between the predictively-coded transmission data and an adjacent row is taken in units of one row of a unit block, the rows are shifted so that the difference value becomes the smallest. Then, by taking the difference and transmitting the shift amount together with the difference value at that time, the coding efficiency can be further improved and the image quality can be further improved as compared with the conventional case.

[Brief description of drawings]

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

FIG. 2 is a block diagram showing an embodiment of a decoding device according to the present invention.

FIG. 3 is a diagram for explaining changes in DCT coefficients due to the discrete cosine transform.

FIG. 4 is a diagram for explaining a calling order of coefficients.

FIG. 5 is a characteristic curve diagram for explaining the average value prediction.

FIG. 6 is a characteristic curve diagram for explaining a quantization / decoding characteristic based on average value prediction.

FIG. 7 is a characteristic curve diagram for explaining the block distortion when the average value prediction is used.

FIG. 8 is a characteristic curve diagram showing a quantization characteristic by adaptive dynamic range coding.

FIG. 9 is a characteristic curve diagram for explaining block distortion when adaptive dynamic range coding is used.

FIG. 10 is a characteristic curve diagram for explaining the edge matching quantization method (1).

FIG. 11 is a characteristic curve diagram for explaining a quantization characteristic by the edge matching quantization method (1).

FIG. 12 is a diagram showing a region in a case where the edge matching quantization method (1) is applied to two-dimensional encoding.

FIG. 13 is a characteristic curve diagram for explaining an edge matching quantization method (2).

FIG. 14 is a diagram for explaining a coefficient call path by DPCM.

FIG. 15 is a diagram for explaining a variable shift DPCM process.

FIG. 16 is a diagram for explaining variable length coding.

FIG. 17 is a diagram for explaining variable length coding.

FIG. 18 is a diagram for explaining a DPCM code.

FIG. 19 is a diagram showing a relationship between edges and DCT coefficients.

FIG. 20 is a connection diagram showing a configuration of a variable shift DPCM decoder.

FIG. 21 is a timing chart used for explaining the operation.

FIG. 22 is a chart for explaining a shift amount.

[Explanation of symbols]

1 ... Image data transmission system, 2A ... Encoding device,
2B ... Decoding device, 5 ... DCT processing unit, 6 ... Predictive coding processing unit, 7 ... DCT / predictive coding determination circuit, 8
Prediction circuit, 10 ... VLC circuit, 11, 31 ... Buffer circuit, 12, 14, 36, 39 ... Inverse quantization circuit, 15 ... Inverse prediction coding circuit, 20 ... DCT circuit,
21, 24 ... Quantization circuit, 22, 35 ... Delay circuit,
23 ... Predictive coding circuit, 25 ... DPCM circuit, 37
...... Inverse DCT circuit, 38 ...... Inverse DPCM circuit, 40 ......
Inverse predictive coding circuit, 50 ... Variable shift DPCM decoder circuit, 51 ... Shift register, 52, 53, 54 ...
... Flip Flop, 55, 59 ... D-Flip Flop.

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] September 4, 1992

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be amended] Title of invention

[Correction method] Change

[Correction content]

Title: Video coding / decoding device

[Procedure Amendment 2]

[Document name to be amended] Statement

[Name of item to be corrected] Claim 1

[Correction method] Change

[Correction content]

[Procedure 3]

[Document name to be amended] Statement

[Name of item to be corrected] Claim 2

[Correction method] Change

[Correction content]

[Procedure correction 4]

[Document name to be amended] Statement

[Correction target item name] 0007

[Correction method] Change

[Correction content]

[0007]

However, since the predictive coding system compresses only the redundancy of the amplitude of the image signal,
The compression rate is lower than that of the discrete cosine transform method that compresses up to the spatial frequency domain, and there is a problem that high coding efficiency cannot be obtained by simply switching between the discrete cosine transform method and the predictive coding method. ..

[Procedure amendment 5]

[Document name to be amended] Statement

[Correction target item name] 0008

[Correction method] Delete

[Procedure Amendment 6]

[Document name to be amended] Statement

[Correction target item name] 0012

[Correction method] Change

[Correction content]

[0012]

When one image is divided into blocks and is coded in block units, the discrete cosine transform coding and the predictive coding are switched according to the nature of the pattern, and the predictive coding is performed. By moving the pixel data from the initial position to a position separated by a predetermined number of pixels and coding the difference value with the pixel data around the position, the coding efficiency can be improved and the overall image quality is also improved. be able to.

Claims (2)

[Claims] 1. An image is divided into unit blocks corresponding to a plurality of pixels, and pixel data of the pixels is orthogonal-transform coded or predictively coded for each unit block and then transmitted. In the moving picture coding / decoding apparatus, when the pixel data is transmitted by the predictive coding, the first pixel in the unit block is moved to a second position separated from the initial position by a predetermined pixel, Moving image coding, characterized in that the image information of the image is transmitted by being compressed by obtaining the difference value between the second pixel data around the second position and the pixel data of the first pixel. Decoding device. 2. An image is divided into unit blocks corresponding to a plurality of pixels, and pixel data of the pixels is orthogonal-transform coded or predictively coded and transmitted for each unit block. In the moving picture coding / decoding apparatus, when transmitting the pixel data by the predictive coding, the pixels in the unit block are moved by a predetermined pixel in the same row or the same column in row units or column units, By calculating the difference value between the pixel data of the second pixel in the row or column adjacent to the second position after the movement and the pixel data of the pixel, it is possible to compress and transmit the image information of the image. A moving image encoding / decoding device having a feature.