CN102790905B - The code-transferring method H.264/SVC arrived H.264/AVC based on P2PVoD video on-demand system - Google Patents

Abstract

The invention proposes a transcoding method from H.264/SVC to H.264/AVC based on a P2PVoD video-on-demand system, belonging to the field of video transcoding. The present invention adopts SVC self-adaptive transcoding, and uses a splitter to divide SVC files into independent multi-layer files according to layers. Before transcoding, the self-adaptive process of space layer and quality layer is performed according to terminal capabilities, and unnecessary layers are discarded. This transcoding process not only effectively reduces unnecessary time overhead, but also provides the client with video quality that matches its capabilities; the combination of pixel domain and transform domain not only reduces time complexity but also effectively The error drift is controlled; the coding control optimization algorithm based on Lagrangian is adopted. When solving the average distortion function, the absolute error function SAD is used instead of the absolute value mean square function SSD, which simplifies the computational complexity; the GOP-based parallel thread processing mechanism is used. Threads process individually encoded GOPs in parallel, greatly reducing transcoding time.

Description

Transcoding method from H.264/SVC to H.264/AVC based on P2PVoD video-on-demand system

technical field

The invention relates to Internet streaming media technology, in particular to a transcoding method from H.264/SVC to H.264/AVC based on a P2PVoD video-on-demand system.

Background technique

With the rapid development of computer network technology and multimedia technology, people's demand for video services is becoming more and more diverse. The research of video technology has become one of the hot research fields of information technology. In March 2003, JVT, a joint project team of ITU/ISO, released the video coding standard H.264/AVC (Advanced Video Coding). Due to its superior compression, high coding efficiency and good network adaptability, it has been widely used usage of.

In order to solve the requirement of accessing multimedia information with the same content from more and more heterogeneous networks, multimedia terminals, and user needs in the expansion of multimedia applications, H.264/SVC (Scalable Video Coding) is proposed on the basis of AVC. standard. At present, SVC has become one of the research hotspots in the field of video technology in recent years. It not only inherits the good coding efficiency and network adaptability of AVC, but also its scalable performance in the time domain, space domain, and quality better meets the needs of Internet video, Diverse demands for different video services such as mobile wireless video, video surveillance, video telephony, and real-time broadcasting.

The degree of use and recognition of SVC in the industry is far less than that of H.264/AVC, which has long been widely recognized and used in the network and industry. It can be seen that many existing terminals, most of which are AVC terminal devices, only support real-time decoding of AVC and decoding of the basic layer of SVC, and cannot realize real-time decoding of SVC. Although the AVC decoding device can support the decoding of the SVC basic layer and achieve basic playback, all terminals hope to obtain video effects that match their capabilities. If the decoding function of the SVC is added to the decoding terminal again, it will inevitably cause a lot of extra overhead on the existing terminal. In addition, in a wireless network, many wireless devices do not have the processing capability of SVC real-time decoding. This requires some necessary processing measures on the content server side before the video reaches the terminal. One situation is to generate several independent code streams with different resolutions on the server side to meet the needs of end users, but this will inevitably increase the burden on the encoding side and lead to a serious waste of bandwidth. Another method is to perform transcoding on the server side to convert the SVC code stream into an AVC code stream to meet the requirements of the terminal without causing other additional overhead. Obviously, the second method can not only reflect the excellent network adaptability of SVC to provide a good adaptive capability for video transmission, but also adapt to the processing capabilities of existing terminals, and provide different terminals with video quality services that best match their capabilities. .

In the current video transcoding technology from SVC to H.264/AVC, there are brute force transcoding and stream rewriting mechanisms.

In transcoding technology, the most intuitive transcoding method is brute force transcoding (full decoding and full encoding): completely decode the existing code stream and then re-encode according to the requirements of the receiving end. This method can obtain the best rate-distortion effect, but its calculation is complicated, time-consuming, and costly, and it is not suitable for the real-time transcoding requirements of communication equipment.

For fast transcoding from SVC to H.264/AVC, the current existing methods are all based on the code stream rewriting mechanism. More representative ones are the code stream rewriting (AVC Rewriting) mechanism from SVC to H.264/AVC proposed by Segall and the code stream rewriting mechanism based on a specific structure proposed by Hannukesela. At present, the best effect is the SVC-to-AVC code stream mechanism based on CGS (Coarse Grain Scalability, coarse grain size) proposed by Segll in 2007. Now there are two standard proposals of JVT-T061 and JVT-V035. This method changes the syntax and semantics of SVC, uses a custom syntax on the SVC encoding side, and rewrites the code stream for the CGS quality scalable layer. Perform lossless transcoding on the SVC code stream with multiple CGS quality layers and the most basic MGS (Medium Grain Scalabilty, medium granularity) layer to obtain video content matching the target quality layer. It has lower transcoding complexity while increasing the base layer with less code rate. Although this method has the advantages of lossless and low complexity, it can only transcode the quality layer of SVC at present, and cannot realize the transcoding of spatial resolution. In addition, this method modifies the syntax and semantics of SVC at the encoding end, which not only affects the encoding efficiency at the SVC encoding end, but also cannot convert the encoded code stream without considering the rewriting of the SVC code stream.

Another code stream rewriting method from SVC to H.264/AVC is a code stream rewriting mechanism based on a specific structure proposed by Hannukesela. This method proposes two code stream rewriting mechanisms with different structures. The first is that the enhancement layer only includes bidirectional predictive images and skippable images, and the bidirectional predictive images in the enhancement layer can be added to the basic layer to form a single AVC code stream after some simple processing. The other is that the SVC coding structure is similar to simulcasting. Each layer of SVC uses independent coding, and there is no reference and prediction between layers. At the transcoding end, simple judgment is made on each layer of the SVC code stream to decide whether to discard it, and then the AVC code stream can be generated by simply modifying the data header information. Although the transcoding operation of this method is simple, it requires strict control at the encoding end, and can only transcode for the above two specific SVC code stream structures, which has too many limitations.

Contents of the invention

The present invention aims to solve the technical problems existing in the prior art, and particularly innovatively proposes a transcoding method from H.264/SVC to H.264/AVC based on the P2PVoD video-on-demand system, and divides the SVC file into For independent multi-layer files, before transcoding, the adaptive process of spatial layer and quality layer is performed according to the terminal capability, and unnecessary layers are discarded, so that this transcoding process not only effectively reduces unnecessary time overhead, but also provides customers with The end provides video quality that matches its capabilities.

In order to realize the above-mentioned purpose of the present invention, the present invention provides a kind of transcoding method based on the H.264/SVC of P2PVoD video-on-demand system to H.264/AVC, it is characterized in that comprising the following steps:

S1. The server side divides the SVC video stream into multiple layered files, wherein the spatial layer and quality layer of the SVC video stream are stored in the corresponding layered file, and the file name of the layered file is DQ _id =1+16 ×D _id +Q _id , D _id represents the number of spatial layers, Q _id represents the number of quality layers, and the spatial layer is composed of a spatial base layer and a spatial enhancement layer, and the quality layer is composed of a quality base layer and a quality enhancement layer;

S2. The server side encapsulates the relevant information of the SVC video stream layered file into the Torrent file;

S3. The client requests the server to download the SVC video stream and sends the client's processing capability information to the server, wherein the processing capability information includes the client's maximum spatial layer processing capacity and quality layer maximum processing capacity;

S4. According to the maximum processing capacity of the spatial layer and the maximum processing capacity of the quality layer fed back by the client, the server side discards the layers corresponding to the spatial layer and the quality layer that exceed the maximum processing capacity of the spatial layer and the maximum processing capacity of the quality layer in the SVC video stream files, so as to realize the adaptive adjustment of space and quality, and obtain SVC effective layered files;

S5. Perform transcoding and decoding on the SVC effective layered file, so as to obtain reuse information;

S6. Perform transcoding encoding on the reused information, so as to obtain an AVC single-layer video stream.

SVC adaptive transcoding is adopted, and the SVC file is divided into independent multi-layer files by layer using a separator. Before transcoding, the adaptive process of the spatial layer and the quality layer is performed according to the terminal capability, and unnecessary layers are discarded. This transcoding process not only effectively reduces unnecessary time overhead, but also provides the client with video quality that matches its capabilities.

In the step S2, the torrent file records the number of layers of the layered files, the spatial layer D _id and the quality layer Q _id to which each layered file belongs, and the relative offset position of each GOP.

Before the step S3, if the client does not have a torrent file, it needs to request the server to download the torrent file.

In the step S5, the transcoding method combining the pixel domain and the transform domain is used to transcode the SVC effective layered file: for the residual data of the SVC quality base layer and the quality enhancement layer that does not use inter-layer residual prediction Code stream, implement transform domain transcoding;

For all spatial layers and quality enhancement layers of SVC, the code stream using inter-layer residual prediction is implemented to perform pixel domain transcoding. The combination of pixel domain and transform domain not only reduces the time complexity but also effectively controls the error drift.

In the step S6, a coding optimization control algorithm based on Lagrangian is adopted, and two coding modes, intra mode and inter mode, are selected to perform transcoding coding on the above reused information, thereby obtaining an AVC single-layer video stream.

In the intra-frame mode, the absolute error function SAD is used to calculate the average distortion function. When solving the average distortion function, the absolute error function SAD is used instead of the absolute value mean square function SSD, which simplifies the computational complexity.

Before the SVC video stream is input to the server, the SVC encoder sets the GOP characteristics through parameters, and encodes each GOP independently, that is, each GOP contains an IDR frame, and takes the IDR frame as the first frame. It can construct a complete image, and can prohibit subsequent frames from making reference to its previous frame, which effectively cuts off the reference relationship of different GOPs, so that each GOP can be independently decoded during the transcoding process.

In the process of transcoding and decoding, different transcoding threads are started based on the GOP picture group, parallel thread transcoding is performed on the GOP, and after processing, the decoded images are reordered to obtain the final AVC single-layer video stream. Adopting GOP-based parallel thread processing mechanism, using threads to process separately encoded GOPs in parallel, greatly reduces the transcoding time, and the programming is easy to implement, and the program is easy to understand, providing a strong guarantee for real-time transcoding.

Each thread has at least one GOP buffer area, which is used to pre-cache the GOP to be processed by the thread. Each thread waits until the entire GOP is ready before performing thread transcoding, wherein the pre-caching time of the GOP is Delay=N* M/f, N represents the total number of threads, M represents the size of each GOP, and f represents the frame rate of the video to be transcoded.

The fewer the number of threads, the shorter the pre-caching time, and the longer the transcoding time. In the actual transcoding process, an appropriate number of threads should be enabled according to the actual video size and the server's own CPU and memory processing capabilities.

In summary, owing to adopting above-mentioned technical scheme, the beneficial effect of the present invention is:

1. SVC adaptive transcoding is adopted, and the SVC file is divided into independent multi-layer files by layer using a separator. Before transcoding, the adaptive process of the spatial layer and the quality layer is performed according to the terminal capability, and unnecessary layers are discarded. This transcoding process not only effectively reduces unnecessary time overhead, but also provides the client with video quality that matches its capabilities;

2. The combination of pixel domain and transform domain is adopted, which not only reduces the time complexity but also effectively controls the error drift;

3. Using the Lagrangian-based encoding control optimization algorithm, when solving the average distortion function, the absolute error function SAD is used instead of the absolute value mean square function SSD, which simplifies the computational complexity;

4. Adopt GOP-based parallel thread processing mechanism, use threads to process GOP encoded separately, greatly reduce transcoding time, and programming is easy to implement, program comprehension is strong, providing a strong guarantee for real-time transcoding.

Additional aspects and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Description of drawings

The above and/or additional aspects and advantages of the present invention will become apparent and comprehensible from the description of the embodiments in conjunction with the following drawings, wherein:

Fig. 1 is a block flow diagram of the present invention;

Fig. 2 is the concrete flowchart of the present invention;

Fig. 3 is a schematic diagram of GOP parallel thread transcoding;

Fig. 4 is a thread flowchart of GOP parallel processing.

detailed description

Embodiments of the present invention are described in detail below, examples of which are shown in the drawings, wherein the same or similar reference numerals designate the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "connection" should be understood in a broad sense, for example, it can be mechanical connection or electrical connection, or two The internal communication of each element may be directly connected or indirectly connected through an intermediary, and those of ordinary skill in the art may understand the specific meanings of the above terms according to specific situations.

The transcoding method from H.264/SVC to H.264/AVC based on the P2PVoD video-on-demand system includes a transcoding preprocessing process, a transcoding decoding process and a transcoding encoding process, as shown in FIG. 1 . The input of the transcoding scheme is the layer-by-layer SVC slice file processed by the SVC splitter, and the output is the H.264/AVC single-layer code stream. In the transcoding preprocessing part, the adaptive process of SVC space and quality is carried out mainly according to the processing capability of the client. In the transcoding and decoding part, the extraction of transcoding reuse information is mainly completed. AVC re-encoding is performed in the encoding part of the transcoding.

The comprehensive transcoding scheme from SVC to AVC is shown in Figure 2, and the specific steps are as follows:

In the transcoding preprocessing part, the spatial resolution or quality adaptive processing is performed on the SVC layer-by-layer fragmented files processed by the separator, and targeted layers are performed according to the client's processing capability, resolution and quality requirements. Discard, select the resolution and quality layer suitable for the user's H.264/AVC decoder device and all layers below this layer as transcoding input.

In the transcoding decoding part, entropy decoding is performed, and then through the process of inverse quantization, inverse transformation, inter-layer intra prediction and inter-layer motion compensation, the video image information required for transcoding and reconstructing images is obtained and stored. In addition, all reference image information obtained by entropy decoding, including mode information and motion vector information of all macroblocks, will be directly used for transcoding.

In the transcoding part of the video, that is, the reconstruction module of AVC, the method of AVC encoder is continued, and the mixed coding method of transformation and prediction is adopted. In the mode selection of the transcoding encoding, the mode selection and motion information of the transcoding decoding end are fully utilized, and the effective mode selection is used to reduce the complexity of calculation while ensuring the quality as much as possible.

The encoding of H.264/AVC undergoes cosine transform and quantization to produce a set of quantized transform coefficients, which are then entropy encoded, and some header information required for decoding (such as prediction mode quantization parameters, motion Vector, etc.) together form a single-layer AVC stream, which is transmitted and stored through NAL, and decoded and played by the client.

The transcoding preprocessing process consists of the following steps:

A1. On the server side, after the SVC video stream is processed by the SVC splitter, it is divided into multiple separate layered files, wherein the spatial layer and quality layer of the SVC video stream are stored in the corresponding layered file, and the layered file The file name of DQ _id =1+16×D _id +Q _id , where D _id represents the number of spatial layers, and Q _id represents the number of quality layers. Use layered files and use DQ _id to name the layered files, which is convenient for quickly locating the target SVC video stream during the distribution process.

In an embodiment of the present invention, for SVC video basic layered files, D _id =0 and Q _id =0, its file name is DQ _id =1+16×D _id +Q _id =1;

For the first enhanced layered file of SVC video, D _id =0 and Q _id =1, its file name is DQ _id =1+16×0+1=2;

For the second enhanced layered file of SVC video, D _id =1 and Q _id =0, its file name is DQ _id =1+16×1+0=17;

For the third enhanced layered file of SVC video, D _id =1 and Q _id =1, and its file name is DQ _id =1+16×1+1=18.

A2, on the server side, encapsulate the relevant information of the layered file of the SVC video stream into the Torrent file, wherein the Torrent file records the number of layers of the layered file, the spatial layer D _id and the quality layer Q to which each layered file belongs _id and the relative offset position of each GOP (Group of Picture, group of pictures), which is convenient for quickly locating the data to be downloaded during the distribution process.

A3. The client requests the server to download the Torrent file. Of course, if the torrent file already exists on the client, it is not necessary to download the torrent file.

A4. The client requests the server to download the SVC video stream and sends the client's processing capability information to the server, wherein the processing capability information includes the client's maximum spatial layer processing capacity and quality layer maximum processing capacity.

A5. According to the maximum processing capacity of the spatial layer and the maximum processing capacity of the quality layer fed back by the client, the server discards the layers corresponding to the spatial layer and the quality layer that exceed the maximum processing capacity of the spatial layer and the maximum processing capacity of the quality layer in the SVC video stream files, so as to realize the adaptive adjustment of space and quality, and obtain SVC effectively layered files. The adaptive adjustment process of space and quality is similar to the SVC video stream adaptive extraction process carried out before SVC video stream transcoding in order to reduce the complexity of transcoding, the difference is that the present invention first divides the SVC video stream into multiple layered files , just discard the unnecessary layered files directly before transcoding the SVC video stream, instead of adaptively extracting the SVC video stream before transcoding the SVC video stream, thus not only reducing the transcoding complexity, but also shortening the transcoding time, Improved transcoding efficiency.

In the embodiment of the present invention, the SVC video stream on the server side includes 3 spatial layers of different sizes and resolutions and 5 quality layers of different quality sizes, while the client's H.264/AVC decoder can support processing 2 spatial layers at most layer and 3 quality layers (that is to say, the maximum processing capacity of the spatial layer of the client is 2, and the maximum processing capacity of the quality layer is 3), at this time, the server directly converts the layer corresponding to the third spatial layer in the SVC video stream Files and layered files corresponding to the fourth and fifth quality layers are discarded.

The transcoding decoding process consists of the following steps:

B1. Perform transcoding and decoding on the above SVC effective layered file, so as to obtain reuse information.

Currently, common transcoding methods include pixel domain transcoding (Spatial Domain Transcoding Architecture, SDTA) and transform domain transcoding (Frequency-Domain Transcoding Architecture, FDTA). SDTA needs to decode the code stream into the pixel domain and then rewrite the code. The computational complexity is higher than that of FDTA, but there is no error drift, and the rate-distortion performance is better; while FDTA only needs to decode the code stream into the transform domain, which reduces the transcoding. The complexity, but the nonlinear budget introduced by H.264/SVC will cause error drift, and the rate-distortion performance is not as good as SDTA.

Therefore, in the present invention, the problem of video quality degradation caused by error drift caused by non-linear operations such as integer cosine transform (Integer Digital Consine Transform, IDCT) deblocking filter (Deblocking Filter, DF) will be considered, and the error drift is carried out. For effective control, it is also necessary to consider the complexity of transcoding time to save encoding time as much as possible.

In order to achieve a better compromise between rate-distortion performance and complexity, the present invention adopts a transcoding scheme combining the pixel domain and the transform domain: for the residual data of the SVC quality base layer and the quality enhancement layer, no inter-layer residual data is used. The difference-predicted code stream implements transform domain transcoding, including entropy decoding and inverse quantization. It does not need to perform inverse transformation process after inverse quantization at the transcoding decoding end. Similarly, it does not need to perform transformation process at the transcoding encoding end. Effective The encoding time is reduced, and since there is no enhanced inter-layer prediction in this part of the code stream, the error drift between layers is also effectively controlled.

For all the spatial layers of SVC (including the spatial base layer and the spatial enhancement layer) and the quality enhancement layer, the inter-layer residual prediction code stream is used, and the pixel domain transcoding is implemented, which effectively controls the error offset and enhances the transcoding performance. rate-distortion performance.

The process of transcoding encoding consists of the following steps:

C1. The Lagrangian-based coding optimization control algorithm performs transcoding coding control on the above reused information, and then obtains an AVC single-layer video stream through normal AVC coding processes such as transformation, quantization, reordering, and entropy decoding. Since H.264 adopts a hybrid coding mode, the above reuse information selects two coding modes, the intra mode and the inter mode. The encoder selects the encoding mode for each macroblock by minimizing the Lagrangian cost function. The Lagrangian cost function of a macroblock is J _MODE (S _k , I _k , |Q, λ _MODE )=D _REC (S _k , I _k , |Q)-λ _MODE ×R _REC (S _k , I _k , |Q )

where S represents an image sequence, and the image sequence S is divided into K different image blocks A _k , the pixel value corresponding to the image block A _k is denoted as S _k , the coding mode corresponding to the image block A _k is denoted as I _k , and the pixel The encoding mode corresponding to the value S _k is Q _k . , λ _MODE represents the Lagrangian operator related to the quantization parameter, and its calculation formula is

λ _MODE = 0.85×2 ^(Q-12)/3

In digital video compression coding, the commonly used average distortion function D _REC (S _k , I _k , |Q) is determined by the original pixels and reconstructed pixels of the macroblock. Generally, there are three calculation methods: the absolute error sum function SAD of the difference, the absolute value square sum function SSD and the mean square error function MSE. In the present invention, in order to simplify the algorithm in mode selection, the SAD algorithm is used to replace the SSD algorithm used in traditional coding.

In intra mode, the SAD formula is as follows:

$\mathrm{<span\; class="notranslate">\; SAD</span>}<span\; class="notranslate">\; =</span>\underset{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; A</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; ]</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; ]</span><span\; class="notranslate">\; |</span>$

Where s[x, y, t] is the current video coded image, and c[x, y, t] is the reference frame data.

In Inter mode, the Lagrangian cost function is more complex than Intra mode due to block-based motion estimation. This scheme calculates the minimum matching error between the image block to be encoded and the reference image block through the Lagrangian cost function (block matching error function), selects the reference block with the minimum matching error, and takes the displacement between the two blocks as the motion estimation motion vector blocks. For a block s(x, y, t) in inter-frame coding mode, its motion vector is mv, given the Lagrangian parameter λ _motion and the reference image c, its cost function minimization model is:

J _motion (s, λ _motion ) = D _motion (s, c) - λ _motion × R _motion (s, c)

Among them, λ _motion is a Lagrangian parameter whose value is R _motion (s, c) is the number of bits of the motion vector. The distortion degree D _motion (S _k , I _k , |Q) is also solved by the absolute value error SSD formula, and the specific formula is as follows:

$\mathrm{<span\; class="notranslate">\; SAD</span>}<span\; class="notranslate">\; =</span>\underset{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; A</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; ]</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; mv</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; mv</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; mv</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; ]</span><span\; class="notranslate">\; |</span>$

Among them, s[x, y, t] represents the current video coded image, c represents the reference frame data, and mv represents the direct motion vector of these two blocks.

In order to meet the higher real-time requirements in the P2PVoD video-on-demand system, the present invention adopts a GOP-level graphic group parallel transcoding processing scheme, as shown in Figure 3, when generating an SVC encoded file, the SVC encoder sets the GOP characteristics through parameters , each GOP is encoded independently, that is, each GOP contains an IDR frame, and the IDR frame is used as the first frame. The IDR frame can not only independently reconstruct the complete image, but also prohibit the subsequent frames from making reference to the previous frame. In this way It effectively truncates the reference relationship of different GOPs, so that each GOP can be independently decoded during the transcoding process.

The process flow of the GOP parallel transcoding processing mechanism is shown in Figure 4. After the server performs space/quality adaptation on the SVC layered file, it locates the video sequence that needs to be transcoded. Since each GOP picture group is coded separately, in the process of transcoding and decoding, GOP can also perform separate internal transcoding. Therefore, decoding can use GOP as a unit to start different transcoding threads, perform parallel transcoding processing on GOP, and then reorder the decoded images after processing to obtain the final AVC video data stream. It can be seen from the figure that each thread must wait until the entire GOP is ready before performing thread transcoding. This requires each thread to have at least one GOP buffer area to store the GOPs to be processed by the thread, and from the perspective of the entire parallel solution, this pre-caching also takes time, which causes a certain delay. For the specific delay time, refer to the following formula:

Delay=N*M/f

Among them, Delay represents the pre-caching time required by the entire GOP parallel scheme; N represents the total number of threads; M represents the size of each GOP; f represents the frame rate of the video to be transcoded. For example, the program starts two threads, the GOP size is 8, and the video frame rate is 30fps, then the pre-cache time Delay=2*8/30=0.5333 seconds. From this, it can be seen that the smaller the number of threads, the shorter the pre-caching time, but the fewer threads, the longer the transcoding time. Therefore, in actual operation, adaptive adjustments can be made according to the size of the actual video, and an appropriate number of threads can be enabled to achieve the best compromise between the pre-caching time and the transcoding processing time, so that the entire parallel transcoding process can reach the optimum time performance. In addition, the processing power of the server's own CPU and memory must also be considered. If too many threads are opened, the server will be overloaded and the transcoding efficiency will decrease instead.

In the description of this specification, descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific examples", or "some examples" mean that specific features described in connection with the embodiment or example , structure, material or characteristic is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Although the embodiments of the present invention have been shown and described, those skilled in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the invention is defined by the claims and their equivalents.

Claims (5)

1. one kind based on P2PVoD video on-demand system H.264/SVC arrive H.264/AVC turn Code method, it is characterised in that comprise the following steps:

SVC video flowing is divided into multiple hierarchical file, wherein this SVC video by S1, server end The space layer of stream and quality layers are deposited to corresponding hierarchical file, the file of this hierarchical file Entitled DQ_id=1+16 × D_id+Q_id, D_idThe representation space number of plies, Q_idRepresent the quality number of plies, and This space layer is made up of spatial base layer and spatial enhancement layer, this quality layers by quality base layer and Quality enhancement layer forms；

The relevant information of this SVC video flowing hierarchical file is encapsulated to Torrent literary composition by S2, server end In part；

The request of S3, client downloads SVC video flowing and by the disposal ability of client Information is sent to server end, and wherein this disposal ability information includes that the space layer of client is maximum Treating capacity and quality layers maximum throughput, before SVC video is streamed to server end, SVC encoder sets GOP characteristic by parameter, by each GOP absolute coding, the most often Individual GOP comprises an every IDR frame, and using every IDR frame as first frame, every IDR frame not only can Independent reconstruct complete image, and can forbid that frame thereafter makes reference to oneself previous frame, so have Block to effect the referring-to relation of different GOP, each GOP can be entered in transcoding process The decoding that row is independent；

Space layer maximum throughput that S4, server end return according to client feedback and quality layers maximum Reason amount, abandons in SVC video flowing beyond this space layer maximum throughput and quality layers maximum The hierarchical file that the space layer of reason amount is corresponding with quality layers, thus realize the adaptive of space and quality Should regulate, it is thus achieved that the effective hierarchical file of SVC；

S5, hierarchical file effective to this SVC carry out transcoding decoding, thus obtain reuse information, adopt The transcoding mode hierarchical file effective to this SVC combined with pixel domain and transform domain carries out turning Code: residual error data and quality enhancement layer for SVC quality base layer are provided without inter-layer residue The code stream of prediction, carries out transform domain transcoding；Strengthen for all space layer of SVC and quality Layer have employed the code stream of inter-layer residue prediction, carries out pixel domain code conversion；

S6, this reuse information is carried out converting coding, thus obtain AVC single layer video stream；

Wherein, during transcoding decodes, use in units of GOP figure group, start different turning Code thread, carries out parallel thread transcoding to GOP, is decoded the weight of image after process again Sequence, obtains last AVC single layer video stream；

The buffer area of each thread at least GOP, treats this thread process for pre-cache GOP, each thread has waited until that whole GOP just carries out thread transcoding after being ready to, wherein The pre-cache time of this GOP is Delay=N M/f, and N represents total Thread Count, and M represents The size size of each GOP, f represents the frame per second treating transcoded video；

Thread Count is the fewest, and the pre-cache time is the shortest, and transcoding time-triggered protocol is the longest, at actual transcoding During should be according to the size of actual video and the CPU of server self and the process of internal memory Ability, opens suitable Thread Count.

The most according to claim 1 based on P2PVoD video on-demand system H.264/SVC To code-transferring method H.264/AVC, it is characterised in that: Torrent literary composition in described step S2 Part have recorded space layer D belonging to the number of plies of hierarchical file, each hierarchical file_idWith quality layers Q_id And the relative deviation post of each GOP.

The most according to claim 1 based on P2PVoD video on-demand system H.264/SVC To code-transferring method H.264/AVC, it is characterised in that: before described step S3, if There is not Torrent file in this client, in addition it is also necessary to downloads this Torrent to server end request File.

The most according to claim 1 based on P2PVoD video on-demand system H.264/SVC To code-transferring method H.264/AVC, it is characterised in that: use in described step S6 based on The code optimization control algolithm of Lagrangian, selects frame mode and inter-frame mode two kinds coding Pattern carries out converting coding to above-mentioned reuse information, thus obtains AVC single layer video stream.

The most according to claim 4 based on P2PVoD video on-demand system H.264/SVC To code-transferring method H.264/AVC, it is characterised in that: in intra mode, put down asking for Absolute error function SAD is all selected to calculate during distortion function.