JP2795225B2 - Video coding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving picture coding method, and more particularly to a moving picture coding method by forcibly generating an uncoded block.

[0002]

2. Description of the Related Art As an image signal data compressing means, an inter-frame encoding method is employed. Although various inter-frame compression coding methods have been proposed, encoding a frame difference of an image signal and orthogonal transform that efficiently uses a correlation in a two-dimensional space of the image with respect to the difference signal are further performed. The transform coding used is often used. In particular, an encoding method based on DCT as orthogonal transform is ISO / IEC 1117 which is a moving image coding standard for storage media.
2-2 (hereinafter referred to as MPEG1) and other international standard encoding methods.

[0003] When performing compression encoding based on the MPEG1 standard, the compression encoding unit is divided into hierarchies. FIG.
As shown in FIG. 1, an input signal is separated into a luminance signal and a chrominance signal, and an intra-coded frame (I frame), a forward predicted coded frame (P frame), or a bidirectional predicted coded frame is separated for each frame. (B frame) prediction method. Further, the frame is divided into units called lattice-like macroblocks, and motion compensation and quantization steps are determined in units of the macroblocks. DCT processing, quantization processing, and variable length encoding processing further divide the luminance signal into four parts. Y0, Y1, Y2, Y in block units
The encoding process is performed in the order of 3, Cb, and Cr.

[0004] Encoding of an image by a conventional DCT-based encoding method will be described by taking MPEG1 as an example. FIG. 8 is a block diagram of an example of a conventional image encoding device. In this conventional image encoding device, input image data is divided into two-dimensional macroblocks, and a motion detection unit 87 detects a motion vector of the macroblock between one and two reference frames. Based on the motion vector information of the macroblock, difference data between motion compensated frames is obtained, the difference data is divided into 8 × 8 blocks, and orthogonal transform is performed in the DCT unit 81. The sampling section 82 performs sampling, and the variable length coding section 83 creates compressed coded data. Based on the code amount of the encoded data output from the variable-length encoding unit 83, the code amount control unit 88 adjusts the quantization coefficient so as to match the target code amount of the compressed data.

In the inter-frame coding method, in order to perform inter-frame prediction, it is necessary to create reference frame data for inter-frame coding of a succeeding frame. When the encoding target frame is an I frame or a P frame, the image data is decompressed by an inverse quantization unit 84 and an inverse DCT unit 85, and the image data is stored in a frame memory 86.
The image data in the frame buffer 86 becomes reference frame data for inter-frame encoding of a subsequent frame.

In such an interframe coding method, when the target frame is an interframe coded frame,
A difference between frames of an image signal is obtained, and the difference value is converted into a variable length code such as a Huffman code. Due to the characteristics of the moving image, the frame correlation after the motion compensation becomes high, the difference value becomes 0 or close to it, and a short variable length code is assigned to them to increase the compression efficiency.

When the data of the divided block after the motion compensation prediction becomes 0, it can be regarded as a non-encoded block, so that the compression efficiency can be further improved.

As a conventional method for improving compression efficiency by creating an uncoded block, there is an image coding control method described in Japanese Patent Application Laid-Open No. 2-241289. Hereinafter, a conventional image encoding control method will be described with reference to FIG.
The subtraction circuit 91, the quantizer 94, the addition circuit 99, and the one-frame delay circuit 98 constitute an inter-frame encoding unit. The valid / invalid block determination circuit 92 accumulates the inter-frame difference signal in block units and determines whether the block is valid or invalid. The difference signal is forcibly set to 0. In this case, when it is determined that the block of the luminance signal is invalid, the block of the color signal corresponding to the block is forcibly processed as invalid, and the luminance is determined. When it is determined that the signal block is valid, the color signal block corresponding to the block is also processed as valid. Therefore, the luminance signal and the chrominance signal are paired and transmitted as an inter-frame coded signal, so that only one of the luminance signal and the chrominance signal is not transmitted.

The above-described valid / invalid block determination is performed using a valid / invalid block determination circuit as shown in FIG. 10
1 is a subtraction circuit for calculating a frame difference value, 102 is an absolute value circuit, 103 is an addition circuit, 104 and 106 are flip-flops constituting a latch circuit, 105 is a comparison circuit for comparing with a threshold, 107, 110 and 111 are selectors, 10
8, 109 are delay circuits. DCK is a data clock signal, BCK is a block clock signal, SEL is a selection control signal added to the timing at the beginning of the block,
YS represents a luminance / color identification signal for distinguishing a luminance signal block from a color signal block, and TH represents a threshold.

The inter-frame difference signal from the subtraction circuit 101 is converted into an absolute value by an absolute value circuit 102. The selector 107 is controlled by the selection control signal SEL to selectively output “0” at the head of each block, and thereafter, to output the contents latched by the flip-flop 104. Is added to the flip-flop 104 by adding 0 to the difference value between the luminance signal Y or the color signals I and Q, and is latched by the flip-flop 104 by the data clock signal DCK. Will be latched. Then, the latch data is added to the addition circuit 103 via the selector 107, so that the difference values in the block are accumulated.

The comparison circuit 105 includes a flip-flop 10
4 is compared with the threshold value TH, for example,
When the latch data is larger than the threshold value TH, a determination signal indicating a valid block is output, and the flip-flop 106 is output according to the block clock signal BCK at the end of each block.
Latched. The selector 110 outputs the luminance / color identification signal Y
Controlled by S, the judgment signal from the flip-flop 106 is selectively output when it indicates the luminance signal block, and selectively outputs the judgment signal via the delay circuit 109 when it indicates the color signal block. The selector 111 is controlled by this determination signal, and when the determination signal indicates a valid block, the difference value via the delay circuit 108 is selectively output, and when the determination signal indicates an invalid block, “0” is selectively output. I do. That is, the difference value of each pixel in the invalid block is forcibly set to 0.

For example, when it is determined that the luminance signal block YB2 is an invalid block, the difference value between the pixels of the luminance signal block YB2 is added to the selector 111 via the delay circuit 108, but is supplied to the selector 110 via the selector 110. According to the determination signal applied to the selector 111, the selector 1
11 selects and outputs "0", and the difference value of the pixels in the invalid block is forcibly set to 0. The difference value between the pixels of the color signal block IQB2 corresponding to the luminance signal block YB2 is supplied to the selector 11 via the delay circuit 109.
Since the determination signal added to 0 is added to the selector 111, the selector 111 selects and outputs "0", and the color signal block IQB2 is also processed as an invalid block.

In this manner, when either the luminance signal or the chrominance signal block is 0, the other block is forcibly set to 0, thereby eliminating code generation in the target block and suppressing this. By allocating the code amount for other blocks to other blocks, the overall image quality can be improved. In addition, since the variable length encoding process is not required, the compression speed is improved.

[0014]

As described above, an image signal is separated into a luminance signal and a chrominance signal, inter-frame difference encoding is performed, and a difference signal from a previous frame is a luminance signal (or a chrominance signal). It has been proposed that when a block is determined to be invalid, a process of forcibly converting a chrominance signal (or luminance signal) into an invalid block to prevent deterioration of reproduced image quality is performed. However, in this case, it is assumed that an invalid block occurs first. In this method, since a simple frame difference from the immediately preceding frame is obtained, if there is a background area such as a screen used in a video conference or the like, the occurrence of an invalid block can be expected, but the original image is particularly important. When a natural image is used, the frequency of occurrence of invalid blocks in this luminance information is extremely low.

In the compression coding method of MPEG1,
There is a definition of a block (skip macroblock) that does not require generation of variable-length code data. The skipped macroblock is composed of a P frame (a forward prediction coded image) and a B frame.
When the difference macroblock data after motion-compensated inter-frame prediction becomes 0, which is recognized in a frame (bidirectionally coded image), the macroblock data after motion compensation of the reference frame is pasted (memory copy is performed. ) Block. As a result, the amount of codes allocated to other macroblocks can be increased, so that block noise generated due to a quantization error by reducing the quantization coefficient is suppressed, and as a result, image quality is improved. The condition for generating a skipped macroblock in the MPEG1 system also requires the motion vector information of the macroblock coded immediately before.

According to the present invention, in the motion-compensated inter-frame coding, the motion vector information is used for a block which is considered to improve the image quality by using the reference frame data as a skip macroblock as it is, rather than coding the difference block information. And by forcibly generating a skip macroblock by data correction of the block data based on the quantized block data, thereby reducing the generated code amount of the block to 0 and increasing the allocated code amount to another block. To improve the image quality.

[0017]

In order to achieve the above object, a moving picture coding method according to the present invention divides an input picture signal into blocks each having a predetermined number of samples, and collectively codes each block. In the inter-frame coding processing method for performing processing, in order to perform motion-compensated inter-frame prediction, a frame memory that stores at least one frame of an image as a reference frame, and the block is referred to between the reference frame and the block. A motion vector detecting unit that detects a block in a reference frame such that the difference from the frame is the smallest, and for each divided block,
A transform unit that performs a discrete cosine transform on the inter-frame difference signal that has been subjected to the motion compensation prediction, a quantizer that performs a quantization on an output obtained by the discrete cosine transform, and encodes the quantized data. A variable-length encoder, and before the variable-length encoding, for the output of the quantizer, by the motion information of the block and the prediction encoding type of the block,
It is determined whether the differential signal of the block is a valid block to be coded or an invalid block that does not need to be coded. If the block is determined to be invalid, the quantized data is forcibly set to 0. It has a data correction means for correcting.

Further, the moving picture coding method according to the present invention comprises:
Noise reduction for performing pre-processing in the time axis direction, wherein a difference value from an immediately preceding frame is obtained for an input image signal, and a pixel value of the immediately preceding frame is substituted for difference data equal to or less than a first threshold value. A filter, a motion / non-motion judging device for judging a moving area when the difference value is equal to or more than a second threshold, and a motion / non-motion judging device for judging a static area when the difference value is equal to or less than the second threshold; The moving picture coding method according to claim 1, wherein the moving picture coding method according to claim 1, further comprising: a filter for performing band limitation in a frame having different band limitation characteristics in a region.

According to the code amount control method of the present invention, it is determined that the last GOP of the image data to be compressed ends halfway and the frame structure and the target code amount are determined.
Control can be performed so that the code amount is initialized in the OP. Thus, the encoded image can be reproduced without leaving data in the buffer used at the time of decompression.

In the moving picture coding method according to the present invention, the quantized data is corrected and the skipped macroblock is forcibly provided on condition that the motion vector information, the frame type of the target frame and the component value of the quantized data are used. In this case, more code amounts can be allocated to other blocks. This makes it possible to reduce the quantization scale of the macroblock as a whole. That is, the quantization error can be reduced, and image quality degradation represented by block distortion can be suppressed. When a skipped macroblock occurs, the variable-length encoding process in the macroblock can be omitted, and the compression speed can be improved.

Further, by using the noise elimination filter and the band-limiting filter in the frame as preprocessing for coding, it is possible to satisfy the conditions for generating a skipped macroblock and the conditions for forcibly performing data correction on the skipped macroblock. Give a boost.

[0022]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the present invention. In FIG. 1, when the compression target frame is an I frame, luminance information and color information divided into blocks are input to the DCT unit 2 and subjected to orthogonal transformation. The output of the DCT unit is sent to the quantization unit, and division is performed on the quantization scale. In the intra-coded frame, data subjected to Huffman coding by the variable length coding processing unit 5 is output without performing data correction. The output data size is input to the code amount control unit 10. The code amount control unit 10 controls the quantization scale to be relatively large when the code amount is larger than the target code amount so as to match the target code amount at the time of compression, so as to match the target code amount. If the generated code amount is smaller than that, control to reduce the quantization scale is performed. Further, the inverse quantizer 6 multiplies the quantized data by the quantization scale of the macroblock, and then performs an inverse DCT operation, and uses this data as reference frame data for inter-frame encoding of a subsequent image. For writing to the frame memory 8.

Next, when the frame to be compressed is a P frame, first, the frame data written in the frame memory 8 is used as a reference frame, and the motion amount of the macroblock is extracted by the motion vector detecting section 9. The predicted block data to be referenced is output from the frame memory 8 based on the amount. The DCT unit 2 receives the difference between the predicted value and the macroblock to be coded, performs orthogonal transform processing, and performs quantization processing. If it is determined by the data correction unit 4 based on the quantized data that the input data should be forced to be a skipped macroblock, all the quantized data is converted into a block of only 0.

The conditions of the MPEG1 standard skip macroblock in the P frame are defined as follows.

First, all the elements of the 8 × 8 quantized intra-block data are 0. Second, the value of the motion vector of the target macroblock is 0 in both the x and y directions. Third, the coding type of the macroblock is determined to be an interframe coded block.

In the present invention, in the P frame,
The conditions for forcibly correcting data are as follows.

1) When the data in the block after the 8 × 8 quantization is Qij, the maximum value thereof is α or less (α is an integer of 0 or more).

2) The integrated value of the absolute value of Oij in the macroblock is not more than β (β is an integer of 0 or more).

3) The x component and the y component of the motion vector of the macroblock to be coded are both 0.

4) The coding type of the macro block is determined to be an inter-frame coded block.

The procedure for forcibly generating a skipped macroblock will be described with reference to the flowchart of FIG. In this method, first, it is checked whether the current block is an inter-frame coded block (step 21).
If the current block is not an inter-frame coded block, the current block is determined to be a coded macroblock. If the block is an inter-frame coded block, it is checked whether both the x component (v _x ) and the y component (v _y ) of the motion vector of the target block are 0 (step 22). If at least one of the x component and the y component is other than 0, the current macroblock is determined to be a macroblock to be subjected to variable length coding.

If both the x component and the y component are 0,
Subsequently, the value of the 8 × 8 component after quantization is checked (step 23). Here, if the components of the 8 × 8 block after quantization are all 0, it is determined to be a skip macro block. If all are not 0, then it is determined whether or not to correct the 8 × 8 block components after quantization. In this case, first, the maximum value of the absolute value of the 8 × 8 component is obtained, and it is checked whether the maximum value is equal to or less than α (for example, 1 or less) (step 2).
4). Here, when the maximum value exceeds α, the block is not set as a skip macro block, and the variable length coding process is performed. If the maximum value is less than α, then 8 × 8
The sum of the absolute values of the components is examined (step 25). Here, if the sum of the absolute values exceeds β (for example, 10), the block is not set as a skip macro block, and the variable length coding process is performed. If the sum is equal to or smaller than β, the block is forcibly determined to be a skip macroblock, all data of 8 × 8 are cleared to 0, and the variable length coding process is not performed.

Next, when the frame to be compressed is a B frame, the motion vector detecting unit 9 extracts the amount of motion of the macroblock by using the frame data written in the frame memory 8 as a reference frame. The predicted block data to be referenced is output from the frame memory 8 based on the amount. The DCT unit 2 receives the difference between the predicted value and the macroblock to be coded, performs orthogonal transform processing, and performs quantization processing. If it is determined by the data correction unit 4 based on the quantized data that the input data should be forced to be a skipped macroblock, all the quantized data is converted into a block of only 0.

The conditions for the MPEG1 standard skip macroblock in the B frame are defined as follows.

First, all the elements of the 8 × 8 quantized intra-block data are 0. Second, the motion vector value of the target macroblock is the most recent in both the x and y directions. Third, the coding type of the macroblock is the same as the coding type of the most recently coded macroblock. That is, it has been determined.

In the present invention, in the B frame,
The conditions for forcibly correcting data are as follows.

1) Assuming that the data in the block after the 8 × 8 quantization is Qij, the maximum value thereof is α or less (α is an integer of 0 or more).

2) The sum of absolute values of Qij in the macroblock is equal to or smaller than β (β is an integer of 0 or more).

3) The absolute value of the difference between the x component and the y component of the motion vector component of the most recently coded macroblock and the motion vector of the macroblock to be coded is equal to or greater than 0 and equal to or less than the integer γ.

4) It has been determined that the coding type of the macroblock uses the same prediction method as that of the most recently variable-length-coded macroblock.

A procedure for forcibly generating a skipped macroblock will be described with reference to the flowchart of FIG. In this method, first, it is checked whether or not it is determined that the current block to be coded is an inter-frame coded block and that the same prediction method as that of the most recently variable-length coded macroblock is used (step 31).

If the current block is not an inter-frame coded block or has not been determined to use the same prediction method as the most recently variable-length coded macroblock, the current block is a coded macroblock. To decide. If this condition is met, the x and y components of the motion vector of the target block are checked, and the difference between the motion vector component and the motion vector component of the most recently variable-length coded block is calculated. If any of the components exceeds γ (for example, 1), the current macroblock is determined to be a macroblock to be subjected to variable-length coding (step 32).

If the absolute value of the difference is less than or equal to γ in both the x and y components, the value of the 8 × 8 component after quantization is checked (step 34). Here, if the components of the 8 × 8 block after quantization are all 0, it is determined to be a skip macro block. If all are not 0, then it is determined whether or not to correct the 8 × 8 block components after quantization. In this case, first, the maximum value of the absolute value of the 8 × 8 component is obtained, and it is checked whether the maximum value is equal to or less than α (for example, 1 or less) (step 35). Here, when the maximum value exceeds α, the block is not set as a skip macro block, and the variable length coding process is performed. If the maximum value is not more than α, the sum of the absolute values of the 8 × 8 components is checked (step 36).

If the sum of the absolute values exceeds β (for example, 10), the block is subjected to variable-length coding without being a skipped macroblock. If the sum is equal to or smaller than β, it is determined that the block is forcibly set as a skip macroblock, all data of 8 × 8 are cleared to 0, and the variable length encoding process is not performed.

Next, a second embodiment of the present invention will be described with reference to the drawings. FIG. 13 is a block diagram showing a second embodiment of the present invention. In FIG. 13, a pre-processing filter 1 is provided before the input image is output to the DCT unit and the motion vector detection unit. This preprocessing filter is used to remove high frequency components due to band limitation in the spatial and temporal directions when performing high-efficiency compression, and to increase the number of blocks satisfying the correction condition for the skip macroblock in the present invention. is there.

The operation flow of one embodiment of the pre-processing filter according to the present invention will be described with reference to FIG. A frame difference is calculated for each pixel in the frame between the encoding target frame data and the immediately preceding frame (step 41).
A noise removal filter having the characteristics shown in FIG. 5 is applied to the frame difference (step 42). It is determined whether the data after noise removal is a moving area or a static area (step 43). In the example shown in FIG. 5, a difference value (for example, plus or minus 10) that sets the same value for the input and the output is used as the threshold for determining the movement. Here, when the difference value falls within the range of plus or minus 10, it is determined that the region is a static region, and when the difference value is out of the range, it is determined that the region is a moving region. Subsequently, a spatial band limiting filter having different characteristics depending on whether it is determined to be a moving area or not in the above-mentioned motion determination (step 43) is used. This is because, in terms of human visual characteristics, in areas where the image is stopped, the degree of band limitation is more sensitive to deterioration than in the moving area. To make the image blur due to the band compression less noticeable in visual deterioration. For example, in the dynamic region, as shown in FIG. 6, a filter that completely reduces high frequency components in the expression band is used.
In the static region, as shown in FIG. 7, a characteristic is provided such that some high-frequency components in the expression band remain.

The following is an example of the characteristics of the filter shown in FIGS.

H (z) = − (1−α) + α {Z (−1) + 2 + z} / 4 (1) Here, when it is determined that the moving area, α is set to 0, and the static area is determined. At this time, α is set to a value from 0 to 1. As α approaches 1, the degree of band limitation in the high frequency range increases. Here, Z (−1) indicates Z to the −1 power.

[0049]

As described above, according to the moving picture coding method of the present invention, more efficient code amount allocation can be achieved by generating skipped macroblocks using data correction after quantization. Therefore, the image quality of the compressed image can be improved. Since the means for data correction uses the data created in the course of compression and the means for forcibly determining a non-encoded block is simple, it can be realized by software.

Applying the pre-processing filter has the effect of increasing the number of blocks that satisfy the conditions for generating skipped macroblocks, in addition to limiting the band or reducing flicker noise, thereby improving image quality.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a moving picture coding method used in the present invention.

FIG. 2 is a flowchart showing the operation of one embodiment of the data correction processing of the present invention.

FIG. 3 is a flowchart showing the operation of one embodiment of the data correction processing of the present invention.

FIG. 4 is a flowchart showing the operation of the pre-processing filter of the present invention.

FIG. 5 is a characteristic diagram of an embodiment of the noise elimination filter of the present invention.

FIG. 6 is a characteristic diagram of an embodiment of the band limiting filter of the present invention.

FIG. 7 is a characteristic diagram of an embodiment of the band limiting filter of the present invention.

FIG. 8 is a block diagram of a conventional inter-frame encoding process.

FIG. 9 is a block diagram of a conventional inter-frame encoding process.

FIG. 10 is a block diagram of a conventional valid / invalid determination circuit.

FIG. 11 is an explanatory diagram of screen division for describing hierarchical encoding of MPEG1.

FIG. 12 is a block diagram illustrating a second embodiment of the moving picture coding method used in the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Pre-processing filter part 2 DCT part 3 Quantization part 4 Data correction part 5 Variable length encoding part 6 Inverse quantization part 7 Inverse DCT part 8 Frame memory 9 Motion vector detection part 10 Code amount control part 11 Inverse DCT part 31 Judgment Part 32 Data processing part

Claims (1)

(57) [Claims] In an inter-frame encoding method in which an input image signal is divided into blocks by a predetermined number of samples and encoding processing is performed collectively for each block, at least in order to perform motion-compensated inter-frame prediction, A frame memory for storing an image of one frame as a reference frame, and a motion vector detecting means for detecting a block in a reference frame that minimizes a difference between the reference frame and the block between the block and the reference frame A transforming means for performing discrete cosine transform of an inter-frame difference signal subjected to motion compensation prediction for each of the divided blocks, and a quantizer for performing quantization on an output obtained by the discrete cosine transform. A variable-length encoder that encodes the data after quantization, and an output of the quantizer before the variable-length encoding. , The value of the block data after quantization, the motion information of the block and the immediately preceding variable-length coded block, and the prediction method of the block. Or a data correction unit for determining whether the block is an invalid block that does not need to be coded and, when the block is determined to be invalid, forcibly correcting the quantized data to 0 . in method for the input image signal, a difference value between the previous frame
If the absolute value of the difference value is equal to or smaller than a predetermined value,
Input image signal so that the absolute value of the difference value of
A noise reduction filter that corrects the signal value, and a frame difference value of the output of the noise reduction filter is equal to or smaller than a threshold
If it is above, it is determined as a moving area, and if it is less than the threshold, it is stationary
A motion / non-motion judging device for judging an area and a band limiting function for a motion area and a static area based on the motion / motion judgment result
A moving image encoding method, further comprising: a filter for performing band limitation in frames having different genders .