CN103916675A - Low-latency intraframe coding method based on strip division - Google Patents

Abstract

The present invention proposes a low-delay intra-frame encoding method based on strip division, which reduces the delay from the two links of acquisition and encoding. First, the acquisition end collects each frame of image in units of strips, and the size of the strips It is related to video resolution, delay requirements, and coding standards; then, encoding is performed in units of slices. In order to improve prediction accuracy and further reduce delays, each slice is divided into multiple sub-slices, and one of the sub-slices is encoded The interpolated enlarged image of the reconstructed image is used as the predicted image for intra-frame prediction, which can effectively improve the prediction accuracy. After testing, the strip-based low-delay codec structure proposed by the present invention can effectively reduce the inherent delay of the codec system, which can be reduced to 150ms for standard-definition video.

Description

A low-delay intra-frame coding method based on slice division

technical field

The invention relates to a new low-delay intra-frame coding method, in particular to a low-delay intra-frame coding based on slice division, and belongs to the technical field of computer vision. the

Background technique

With the continuous improvement and development of network technology and terminal processing capabilities, in order to continuously improve compression performance, ITU and ISO organizations have introduced a series of video coding standards, including the H.26x series organized by ITU and the MPEG-x series standards organized by ISO. And the latest HEVC standard. The latest HEVC standard is committed to satisfying users' 1) high-definition, 2) 3D, 3) mobile wireless to meet the needs of new fields such as home theater, remote monitoring, digital broadcasting, mobile streaming media, portable video, medical imaging, etc. . There are multiple configuration modes in the HEVC standard, including HE (High Efficiency) high performance and LC (Low Complexity) low complexity configuration. the

Among these coding standards, there are mainly two categories: one is storage-oriented, and the other is transmission-oriented. The H series are mainly transmission-oriented, and the Mpeg series are mainly storage-oriented. Although the video coding standards are formulated by different organizations, the year of completion and the background of the application are different, the basic coding framework adopted is the same, and the basic framework of motion compensation + DCT is mostly used. Video frames under this framework are generally divided into three types: I (Intra-frame), P (Predictive-frame), and B (Bidirectionally predicted-frame): I frames are encoded through processes such as transformation and quantization; The reconstructed image of the encoded frame is used as a reference, and the residual is encoded after motion compensation; the reconstructed image of the B frame is used as a reference, and the residual is encoded after motion compensation. the

Among the three types of video frames, although the number of I frames is relatively small, the number of coding bits per frame is much higher than that of P and B frames, and it also occupies a considerable proportion in the final generated code stream. In H.264 In the standard, the compression ratio of the I/P frame is: I:P=1:3~5, that is, the bit rate of the I frame is 3~5 times that of the P frame, and in the HEVC standard, with the new complex technology With the introduction of , the compression ratio of I frame and P frame is further expanded, and in some videos, it can reach 1:10. For a video stream with a constant output bit rate (CBR), a sudden increase in the bit rate of the I frame will directly reduce the number of coding bits for the subsequent P/B frames, thereby affecting the quality of the restored image and the predicted image. Improving the compression ratio of I frames not only plays a vital role in the stability and continuity of video quality, but also reduces the delay of the codec system. That is, if the code rate of the I frame is too high, more network bandwidth will be occupied to transmit the compressed code stream. the

In the process of video encoding, especially on mobile platforms, when bandwidth and complexity are required, it is hoped that the encoding and decoding algorithms will achieve low latency and low complexity. In the video encoding process, delays are mainly caused by two reasons:

(1) The inherent time of encoding and decoding, if the encoding and decoding speed can reach real-time, then there will be no delay. the

(2) Channel transmission, if the channel is wide enough, there will be no congestion and no delay. the

If the channel is not considered, how to reduce the inherent delay of codec is very important. The delay of a codec system consists of the following parts (see Figure 1), including image information acquisition, compression processing, sending buffer, data link transmission, receiving buffer, decompression, image data format conversion and display, etc. Delay. In the case of real-time encoding with a frame rate of 25 frames per second, the encoder completes the compression process within 40ms after a frame of image acquisition is completed; when the encoded frame type is I-P, the bit ratio of the encoded output is generally 8:3, Therefore, the buffer size is usually set to be 3.5 to 4 times the average code flow per frame to satisfy the buffering of the I-frame code flow. After being transmitted through the data link, the decoding end decompresses the compressed code stream after receiving it, and finally displays it on the display device. In Figure 1, the following two situations are assumed:

1): The data link transmission is fixed throughout the encoding and decoding process (CBR type channel), and the delay is not included. the

2): In the real-time acquisition, processing, and transmission system, the receiving buffer and the decompression process are processed in parallel, and only one time can be calculated. the

Under the standard encoding framework, with the frame as the basic unit, the total delay analysis of the encoding and decoding system is as follows: the acquisition time T1 is the time required to acquire a frame of image, taking the PAL standard as an example, T1=1000/25=40ms. the

Encoding time T2: In the case of real-time encoding, the time to encode one frame of image is: T2=40ms. Bit stream sending time T3: The part that has a greater impact on delay is the sending buffer. The buffer is set to solve the transmission and storage problems of variable bit rate bit streams. The larger the buffer, the more it can withstand fluctuations in the bit rate, but the greater the delay caused, and vice versa. Taking I-frame encoding as an example, the generated code stream may fill 3/4 of the buffer, so the delay of the sending buffer is generally T3=3×40=120ms, and it may lead to 140ms in extreme cases where the buffer is full ~160ms latency. the

Code stream receiving time T4: the time to decompress the first frame of the image and convert the image data format, here recorded as T4=25ms. the

The display time T5 is sent to the display device after decompression, and the time to display a frame of image is T5=40ms. The above time is accumulated, and the total encoding and decoding delay is T=T1+T2+T3+T4+T5=265ms. Among the above-mentioned delays, I frame, as a key frame type, has the largest encoding and decoding delay, and at the same time contributes the largest to the entire encoding and decoding delay. Latency has an important impact. the

Contents of the invention

The purpose of the present invention is to provide a low-delay intra-frame encoding method based on slice division. The method is to analyze the delay of video encoding and decoding, and provide a low-delay encoding structure suitable for I-frame encoding. In this structure, two aspects of the acquisition end and the encoding end will be mainly considered. First, the acquisition end collects each frame of image in units of strips, and the size of the strips is related to video resolution and delay requirements; then, encoding is performed in units of strips, in order to improve prediction accuracy and further reduce delays, each A slice is divided into multiple sub-slices, and one of the sub-slices is used as a prediction unit for intra-frame prediction, which can effectively improve the prediction accuracy. the

In order to achieve the above object, the present invention adopts the following technical solutions. It is characterized in that comprising the following steps:

Step 1: Video acquisition is performed in units of slices, and the input image is divided into several slices for acquisition to reduce the acquisition delay. The determination of the size of the slices is related to the size of the basic coding unit in the adopted video coding standard. Set The number of scanning lines of a frame of image is N _frame , N _bcu indicates the number of pixel lines contained in the basic coding unit, N _{slice_bcu} indicates the number of pixel lines in each slice, and the unit of the slice is a multiple of N _bcu Divide a frame of image into slices, N _{slice_bcu} =α×β×N _bcu , where the parameter α is a delay adjustment parameter, β is a resolution adjustment parameter, and both α and β are positive integers;

Step 2: Carry out predictive encoding in units of strips. In order to improve the prediction accuracy, the strips are extracted by interlacing and/or every other column, and further divided into multiple sub-stripes; the above-mentioned multiple sub-strips are divided into the following two ways Predictive coding: firstly, one of the multiple sub-strips is used as a basic sub-strip, and the prediction method specified in the video coding standard is directly used for coding, and the coded stream directly enters the output buffer and is transmitted to the The decoding end performs decoding; secondly, reconstructs the coded stream of the basic sub-strip, and performs interpolation and amplification processing on the reconstructed image, and enlarges it to the same resolution as the strip, and obtains the interpolated image as other The predicted image of the sub-strip; thirdly, for the remaining several sub-strips, during the encoding process, according to the predicted image, the method of directly calculating the difference of the corresponding position point is used for prediction, and the residual data obtained by encoding; when the image is decoded, The rest of the several sub-strips are restored by using the reconstructed and enlarged images of the basic sub-strips as predicted images, which avoids repeated interpolation processing. the

The above slice division is related to coding standards. For the H.264 coding standard, the basic coding unit is a macroblock, and for the HEVC coding standard, the basic coding unit is a CU or LCU. In order to reduce the code stream difference between the I frame and the P frame, so that the compressed code flow of each frame image is stable, the type of the input image is judged. If the input image is an I frame, the divided strips will all adopt intra-frame coding , if the input image is a P-frame or B-frame image, then one slice in the P-frame or B-frame image adopts the Intra coding mode, and the rest of the slices still use the original P-frame Inter coding. the

The low-delay intra-frame encoding method of slice division provided by the present invention can effectively improve the prediction accuracy and further reduce the encoding delay. the

Description of drawings

Figure 1 is a schematic diagram of system delay;

Figure 2 is the division method of low-delay coding based on slice acquisition;

Figure 3 is the prediction of spatial adjacent points in JVT;

Figure 4 is the intra prediction mode in the HEVC standard;

Figure 5 is the prediction distance of pixels under different prediction methods;

Detailed ways

As mentioned above, according to the low-delay intra-frame coding method of the slice division, the present invention realizes the reduction of the code stream without reducing the image quality, thereby reducing the delay. the

Illustrate the implementation of the present invention below in conjunction with accompanying drawing, in Fig. 1, assume following two situations:

1): The data link transmission is fixed throughout the encoding and decoding process (CBR type channel), and the delay is not included. the

2): In the real-time acquisition, processing, and transmission system, the receiving buffer and the decompression process are processed in parallel, and only one time can be calculated. the

Under the standard encoding framework, with the frame as the basic unit, the total delay analysis of the encoding and decoding system is as follows: the acquisition time T1 is the time required to acquire a frame of image, taking the PAL standard as an example, T1=1000/25=40ms. the

Encoding time T2: In the case of real-time encoding, the time to encode one frame of image is: T2=40ms. Bit stream sending time T3: The part that has a greater impact on delay is the sending buffer. The buffer is set to solve the transmission and storage problems of variable bit rate bit streams. The larger the buffer, the more it can withstand fluctuations in the bit rate, but the greater the delay caused, and vice versa. Taking I-frame encoding as an example, the generated code stream may fill 3/4 of the buffer, so the delay of the sending buffer is generally T3=3×40=120ms, and it may lead to 140ms in extreme cases where the buffer is full ~160ms latency. the

Code stream receiving time T4: the time to decompress the first frame of the picture and convert the image data format, here recorded as T4=25ms. the

The display time T5 is sent to the display device after decompression, and the time to display a frame of image is T5=40ms. The above time is accumulated, and the total encoding and decoding delay is T=T1+T2+T3+T4+T5=265ms. the

Among the above delays, I frame is the key frame type, its codec delay is the largest, and at the same time it contributes the most to the entire codec delay. have an important impact. the

To this end, it is necessary to reduce the delay of encoding from the two links of acquisition and encoding. the

Step 1: Reduce acquisition delay based on stripe division

Figure 2 shows the low-latency encoding division method based on strips to reduce the acquisition delay. The division of the strips should not be too small or too large; if it is too small, it will affect the prediction effect during encoding processing; if it is too large, the acquisition The delay reduction effect is not obvious. Interlaced scanning and progressive scanning will be analyzed separately below. the

Taking an interlaced video sequence with an interlaced scanning of 25 frames per second and a resolution of 720×576 as an example, the number of lines of each image is N _frame =576. For H.264 and its previous standards, the macroblock line N _MB = 16 units will be divided into slices; the HEVC standard handout LCU will be divided into slices, N _MB = 64. Here, H.264 and its previous standards are used for analysis, and the analysis of HEVC standard is similar, so I won’t go into details here. N _MB contains pure odd-numbered lines or pure even-numbered lines, and a frame of image has a total of N _{slice_MB} =N _frame /N _MB =36 strips; for the system delay based on slices: one slice can be collected for encoding, and the acquisition requires The time is: T1=40ms/36=1.11ms. In a slice-based codec, the compressed code flow of a frame of image is relatively stable, the buffer size can be set to 1.25 to 1.5 frame length, and the sending buffer delay is generally T3=40ms; T2, T4, The calculation method of T5 is similar to the frame-based encoding method, which reduces the time by more than 100ms compared with the frame-based encoding and decoding method.

Progressive scanning: Striping can be either progressive or interlaced. However, the present invention adopts the form of interlaced division, and N _MB only includes odd-numbered or even-numbered lines, that is, the first SLICE will be composed of odd-numbered lines such as the first line, the third line, and the fifth line, and the second SLICE will be composed of Even-numbered lines such as the second line, the fourth line, and the sixth line form.

The above two stripe division methods in the progressive scanning method have little difference in delay, and the first method is the same as the interlaced scanning method. Compared with the first method, the second method requires one more band during acquisition, and the time required for acquisition is: T1=(40ms/36)*2=2.22ms. the

Considering that most real-time systems are interlaced scanning, the subsequent implementation analysis will use interlaced scanning to analyze strip division and encoding processing. the

The specific implementation steps of the slice-based coding prediction mode are as follows:

In the standard encoding process, a multi-mode prediction technology is used to predict from multiple angles. The intra-frame prediction algorithm used in the H.264 standard uses a 16×16 macroblock and a 4×4 block as the basic prediction unit, and predicts each point in the block in nine directions. Taking a 4×4 block as an example, as shown in the left figure of Figure 3, a, b, ..., p are the current block to be predicted, and the surrounding 17 points Q, I, ..., P are coded points. For a-p point by point, along the nine directions of 0, 1, ..., 8 and direction 2 (DC prediction) shown in the right figure of Figure 3, take the points in P-Q-H, use the appropriate prediction formula to calculate the predicted value, and the original sampling Values are differentiated, and the mode with the smallest difference is the final prediction mode, and finally DCT coding is performed on the prediction residual coefficients. the

In the latest HEVC standard, three new concepts, CU (Coding Unit), PU (Prediction Unit), and TU (Transform Unit), are introduced. The coding unit is similar to the macroblock concept in H.264/AVC, and its size can be up to 64×64; the prediction unit is the basic unit for prediction; the transformation unit is the basic unit for transformation and quantization. The separation of the three units makes the processing links of transformation, prediction and encoding more flexible and more in line with the texture characteristics of video images. The size of the prediction unit can be 4×4, 8×8, 16×16, 32×32, 64×64, the size of the block is different, and the prediction mode is also different, respectively 17, 34, 34, 34, and 3 Two modes are optional, as shown in Figure 4, the unified intra prediction angle is: +/-[0, 2, 5, 9, 13, 17, 21, 26, 32]/32, the mode selection standard is the same as JVT Same standard. the

These prediction modes have their inherent complexity. The present invention introduces a layered prediction mode based on the nearest pixel as shown in Fig. The reconstructed gray points are used for second-level predictions. And through the concept of average prediction distance (that is, the average value of all current pixels and their reference pixel distances, represented by D _pred ), and use the average distance to analyze the prediction distance of the standard method and the method of the present invention.

Firstly, the first strip collected is divided into four sub-strips, and the extraction is performed in the pattern of alternate rows and columns. Then, one of the sub-strips is coded according to a standard prediction mode, that is, after transformation, quantization, entropy coding and other processing processes, a reconstructed image is obtained. Afterwards, the reconstructed image is enlarged to the size of the strip by using bilinear interpolation method. Finally, the remaining sub-strips will be predicted according to the pattern shown in Figure 5(B). The following is an analysis of the prediction distance of the present invention and the prediction distance of the H.264 standard. Here, only the vertical forecasting mode in the standard is taken as an example for theoretical analysis, and this method can be applied to other forecasting modes in the same way. the

Assuming that the size of the current basic block to be encoded is 2N×2N (usually N=2, 4, 8, 16...), D _{i, j} represents the distance between the pixel with coordinates (i, j) and its reference pixel, the data The average prediction distance D _pred of a block can be obtained according to the following steps.

First calculate the predicted distance D _cot of a single column of pixels:

The total prediction distance D _total of the data block is the sum of the prediction distances of each column:

$\begin{array}{c}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; total</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; col</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; col</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; col</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}\\ <span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; col</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>\\ <span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}\end{array}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

Figure 5(A) is the intra-frame prediction vertical mode of H.264. In this mode, the prediction distance of all pixels in the same row is equal, that is, D _j =i. The number marked on the current pixel point in the figure is the prediction distance of the point , so the average prediction distance D _{pred_A} of all pixels in the data block is:

$\begin{array}{c}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; pred</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; A</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; total</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; A</span>}}<span\; class="notranslate">\; /</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; )</span>\\ <span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; j</span>}}\\ <span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>\\ <span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\end{array}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

(2) The arrow in Figure 5(B) marks the reference pixel point of each pixel, which is the same as in Figure 5(A), the prediction distance of the first-level pixel point is determined by the distance between it and the reference pixel; due to the second-level prediction All points in are predicted from neighbors, so the prediction distance is 1. the

The average prediction distance of the inventive method is calculated as follows: first, the first-level prediction distance D _{level_1} is the sum of N column pixel point prediction distances in the first level:

${\mathrm{D.}}_{\mathrm{level}\_1}={\mathrm{D.}}_{\mathrm{col}\_1}+{\mathrm{D.}}_{\mathrm{col}\_3}+......+{\mathrm{D.}}_{\mathrm{col}\_2N-1}=\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{D.}}_{\mathrm{col}\_j}$ Since the first-level prediction adopts interlaced prediction, the prediction distance of each column of pixels D _{col_j} =1+3+...+2N-1,

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; level</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; col</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; j</span>}}$

Substitute into the above formula to get: =N×(1+3+5+.…+2N-1)=N ³

The second-level prediction distance D _{level_2} is the sum of the prediction distances of the three second-level prediction points respectively corresponding to all the points obtained from the first-level prediction (2N in total), that is, D _{level_2} =2N×3.

Therefore, the prediction distance D _{total_B} of the entire prediction pixel block using the method shown in Figure B is:

D _{total_B} =D _{level_1} +D _{level_2} =N ³ +6N

D _{pred_B} =D _{total_B} /(2N×2N)

The average prediction distance D _{pred_B} is: =(N ³ +6N)/4N ² (3)

(3) Let λ be the ratio of D _{pred_A} and D _{pred_B} , we can get:

$\mathrm{<span\; class="notranslate">\; \&lambda;</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; pred</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; A</span>}}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; pred</span>}<span\; class="notranslate">\; \_</span>\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}{{\mathrm{<span\; class="notranslate">\; N</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 6</span>}}$

The above formula shows that there is a gap of several times between D _{pred_A} and D _{pred_B} , and the larger the value of N, the larger the gap.

Comparing formulas (2) and (3), it can be seen that the hierarchical prediction method greatly reduces the average prediction distance. For a prediction block of 4x4 size (N=2), D _{pred_A} = 2.5, D _{pred_B} = 1.25 can be calculated, and the difference between the two is doubled; for a larger prediction block of 16x16 (N=8), the average of the two methods The prediction distances are 8.5 and 2.2 respectively. Compared with the H.264 method, the average prediction distance of the hierarchical prediction method based on the nearest pixel is reduced by nearly 75%, and the advantage is more obvious.

performance analysis

The H.264 standard baseline profile is implemented and tested by the method of the invention and compared with the standard algorithm results. For the restored image quality, the present invention still uses the traditional peak signal-to-noise ratio (PSNR) as a measure. By adjusting the quantization parameters, the corresponding coding bits (Kb/frame) of each sequence are respectively measured when the standard algorithm and the present invention have the same PSNR (dB) value. For the test sequence, the present invention selects the violently moving coastguard, the relatively static mother, and the complex-textured flower from the Cif sequence. The test results are shown in Table 1. the

Table 1 Algorithm of the present invention and standard algorithm test result comparison

Claims (4)

1. A low-delay intraframe coding method based on slice division, characterized in that: Step 1: Video acquisition is performed in units of slices, and the input image is divided into several slices for acquisition to reduce the acquisition delay. The determination of the size of the slices is related to the size of the basic coding unit in the adopted video coding standard. Set The number of scanning lines of a frame of image is N _frame , N _bcu indicates the number of pixel lines contained in the basic coding unit, N _{slice_bcu} indicates the number of pixel lines in each slice, and the unit of the slice is a multiple of N _bcu Divide a frame of image into slices, N _{slice_bcu} =α×β×N _bcu , where the parameter α is a delay adjustment parameter, β is a resolution adjustment parameter, and both α and β are positive integers; Step 2: Carry out predictive encoding in units of strips. In order to improve the prediction accuracy, the strips are extracted by interlacing and/or every other column, and further divided into multiple sub-stripes; the above-mentioned multiple sub-strips are divided into the following two ways Perform predictive coding: first, use one of the multiple sub-strips as a basic sub-strip, directly use the prediction method specified in the video coding standard for coding, and the coded stream directly enters the output buffer and is transmitted to The decoding end performs decoding; secondly, reconstructs the coded stream of the basic sub-strip, and performs interpolation and amplification processing on the reconstructed image, and enlarges it to the same resolution as the strip, and obtains the interpolated image as other The predicted image of the sub-strip; thirdly, for the remaining several sub-strips, during the encoding process, according to the predicted image, the method of directly calculating the difference of the corresponding position point is used for prediction, and the residual data obtained by encoding; when the image is decoded, The rest of the several sub-strips are restored by using the reconstructed and enlarged images of the basic sub-strips as predicted images, which avoids repeated interpolation processing. 2. The intra-frame coding method according to claim 1, wherein: the slice division is related to a coding standard, for the H.264 coding standard, the basic coding unit is a macroblock, and for the HEVC coding standard, the basic The coding unit is CU or LCU. 3. intra-frame coding method as claimed in claim 1, is characterized in that: for reducing the code flow difference between I frame and P frame, make the compressed code flow of every frame image steady, the type of input image is judged , if the input image is an I frame, the divided slices will all adopt intra-frame encoding; if the input image is a P frame or B frame image, then one slice in the P frame or B frame image will use Intra encoding, and the remaining slices The belt still adopts the Inter coding of the original P frame. 4. The intra-frame coding method according to claim 1, wherein the coding is performed after a slice is collected, so as to reduce the collection delay.