TWI433543B - Distributed coding for wireless high-speed video codec system - Google Patents

Description

High-speed encoding wireless distributed video codec system

The present invention relates to a distributed video codec system, and more particularly to a wireless distributed video codec system for high speed coding in a wireless communication environment.

Digital video encoding and decoding methods vary depending on the application. Early video encoding and decoding methods are mainly used in entertainment applications or network and storage applications (such as multicast, Blu-Ray, etc.). Such applications are designed for decoding methods. It is more streamlined than the coding method for users to use a simple and low-cost decoding device.

Recently, digital video applications have changed, such as today's wireless cameras, wireless low-power monitors, mobile file scanners, mobile video conferencing, mobile video e-mail, disposable cameras, visual sensor networks, webcams, distributed video strings. Streaming, multi-view video entertainment, wireless capsule endoscopy, etc., on the contrary, the encoding method for generating digital video data ends is more streamlined than the decoding method to reduce the cost of such devices. The biggest difference between the decentralized video coding method and the related standard of the early video coding method is the execution of the Motion Estimation program. Generally, the coding end in the early coding standard eliminates the temporal redundancy. In order to achieve the compression effect, the calculation amount is mostly concentrated on the encoding end, and the recent decentralized coding method concentrates the calculation amount on the decoding end. Therefore, the coding end of the distributed video codec system contributes to the above products. development of.

Since most of the products listed above are used in the wireless communication environment, the reliability of the power or transmission quality must be balanced. Therefore, the encoding end of the distributed video codec system usually requires the following conditions:

(1) The computational complexity burden is concentrated on the decoding end, so that the various costs of the encoding and decoding end, such as: calculation amount, power consumption, and more balance, can also achieve the low power (low power) and low complexity of the coding end. .

(2) The rate-distortion efficiency can be close to the codec of the current encoder motion estimation architecture, which can adapt to the wireless transmission environment with limited bandwidth, so there is the development of Distributed Video Coding and DVC.

The following is a further introduction to a Wyner-Ziv distributed video codec system. Please refer to the seventh figure. The encoder (30) includes a quantizer (31) for acquiring digital video data. A similar information word is represented by a code word to achieve the first compression purpose; and a Slepian-Wolf encoder (32) is connected to the quantizer (31) to obtain the first compressed video. Data, and according to the generation of parity bits, and the Slepian-Wolf encoder (32) transmits only some of the parity bits according to the demand, and the size of the transmitted parity bits is less than or equal to the entropy coding of the data source. Entropy Code Length, which achieves the second function of compressing information.

The decoding end (40) of the Wyner-Ziv distributed video codec system includes a Slepian-Wolf decoder (41) and a coefficient reconstruction module (42), wherein the Slepian-Wolf decoder (41) is linked. Up to the Slepian-Wolf encoder (32) to obtain the second compressed video data, and then cooperate with the coefficient reconstruction module to receive the side information Y (Side Information) Y to restore the correct video data X; wherein the side decoding information Y can be generated by interpolation or extrapolation, wherein the interpolation method can use the average interpolation method or the dynamic compensation interpolation method, and the extrapolation method can be the dynamic compensation method, and the quality is good by the side. Decoding the information Y can reduce the size of H(X/Y), which means that the generation of the number of parity bits can be reduced. It can be seen that the quality of the side decoding information is related to the correctness of the restored video data.

However, the simple architecture of the above Wyner-Ziv distributed video codec system does not consider the problem of video data transmission. As mentioned above, the distributed video codec system is most commonly used in wireless communication environments; An improved distributed video codec system is provided. Please refer to the figure 8 in the figure, which is an architecture diagram in a pixel domain, which includes an intra-frame coding module (50) and an inter-picture decoding module. (60), wherein the intra-coded single module (50) comprises: a quantizer (51) for acquiring a Wyner-Ziv picture (S) in the video data for quantization and compression, and the quantizer (51) Is a 2 ^M level scaler quantizer; a high speed encoder (52) is connected to the quantizer (51) to obtain a quantized compressed picture and quickly encode it; a buffer (53), It is connected to the output end of the high speed encoder (52); and an intra picture coding unit (51) obtains the main picture (K) in the video material and directly performs fast coding.

The inter-picture decoding module (60) includes: a high speed decoder (61) connected to the high speed encoder (52) through the buffer (53) to obtain a coded Wyner-Ziv picture and a parity bit. a reconstruction unit (62) is connected to the high speed decoder (61); an intra-picture decoding unit (63) is connected to the intra-picture coding unit (54), and is decoded to output a main picture; The decoding information generating unit (64) is connected between the in-picture decoding unit (63) and the reconstruction unit (62), and is interpolated or externally according to the decoding main picture (S') output by the in-picture decoding unit (63). Interpolation produces side decoding information ( ), then decode the information ( The output to the high speed decoder (61) and the reconstruction unit (62) causes the reconstruction unit (62) to restore a high quality Wyner-Ziv picture.

The above system architecture facilitates the transmission of the wireless communication environment. The reason is that, as shown in the system diagram of the system of the discrete cosine transform (hereinafter referred to as DCT conversion) in the ninth diagram, the Wyner-Ziv picture pre-defined at the encoding end (50) ( S) After DCT conversion, it is converted into the DCT domain, and then its transform coefficients are divided into several coefficient bands, which are independently coded by a 2 ^M level scaler quantizer, that is, fixed length The two-bit word code is added as a representative quantized symbol (fixed-length binary code words), and then transmitted.

As for the decoding end (60), the in-picture decoder decodes the encoded main picture into K ', and then generates the interpolation by the side information generating unit. And performing a blockwise DCT with the Wyner-Ziv picture, that is, dividing the conversion coefficient into several corresponding coefficient bands to generate side information. On the other hand, the high speed decoder (61) sequentially decodes other bit planes from the most significant bit-plane to obtain corresponding parity bits and side information, thereby solving the current The associated bit plane, if the high speed decoder (61) cannot solve the bit planes reliably, the additional parity bits can be requested through the feedback channel, repeating this step until an acceptable one occurs. After the bit rate is misaligned, the current bit plane can be obtained, and the quantized symbols are inverse quantized and discrete coherently transformed, and finally the Wyner-Ziv picture (S') is reconstructed. Yes, this decentralized video editing The decoding system can facilitate transmission and reception in wireless communication by the above encoding and decoding methods.

As can be seen from the above two types of distributed video coding systems, the calculation amount of the coding end (50) is much lower than that of the decoding end (60), so the codec complexity of the video data is often evaluated by the execution time. Recently by Instituto Superior The (IST) imaging team launched a 27-month research project to propose a new DISCOVER software for coding efficiency, error recovery, adjustability, and model-based video coding. Breakthroughs have paved the way for next-generation video coding.

However, whether the encoder in the distributed video coding system can achieve the above intended purpose by using a DISCOVER software, the following analysis data is further provided for comparison. That is to say, the statistical execution of the DISCOVER software encoder and other video standard encoders to perform the encoding time, although the DISCOVER encoder time is less than other encoders, it is not difficult to find that the encoding time of the main picture accounts for almost the entire encoder execution time; Therefore, if the defect can be effectively improved, the coding speed can be effectively improved, and the future development of the distributed video coding system can be helped.

The main object of the present invention is to provide a high-speed encoded wireless distributed video codec system, which has a high-speed encoding function of a main picture and improves the overall coding rate.

The main technical means for achieving the above purpose is that the high-speed encoded wireless distributed video codec system includes an intra-picture coding module and an inter-picture decoding module, wherein: the internal picture coding module includes: A non-main picture coding unit encodes a non-main picture in the video data; a primary sampling unit performs sub-sampling on the main picture in the video data to generate a plurality of sample symbols; and a multiple description coding unit, The system includes a plurality of intra-frame encoders and is commonly connected to the sub-sampling unit to obtain a quantized main picture composed of a plurality of sampled symbols, and then disassembled into a plurality of low-resolution pictures, and corresponding intra-picture coding The device performs encoding and then transmits it to the associated channel. The low-resolution picture transmitted by the encoder channel in the complex picture must have similar but independent statistical characteristics. The inter-picture decoding module includes: a non-main picture decoding mode. a group, linked to the non-home picture coding unit, to decode a non-home picture; a multiple description decoding unit, including a plurality of intra-picture decoders, respectively coupled to the corresponding intra-picture encoder to restore the low-resolution picture of the corresponding channel; and a super-resolution restoration reconstruction unit connected to the multi-picture decoding unit of the multiple description decoding unit The complex low-resolution screen combination is restored to the original main screen.

The above invention mainly provides a multiple description decoding unit for the main picture coding part of the intra picture coding module, and further divides the quantized main picture after the second acquisition into a plurality of low resolution pictures, and respectively generates low Corresponding channel transmission of the resolution picture; in the present invention, the multiple description decoding unit can distribute the encoding and transmission of a single main picture, and complete the encoding and transmission of the single main picture at a faster speed.

Referring to the first figure, the wireless distributed video codec system includes an intra picture coding module (10) and an inter picture decoding module (20), wherein: the internal picture coding module (10) includes There is: a non-main picture coding unit (11), which encodes a non-main picture (such as a Wyner-Ziv picture) in the video data; a sampling unit (12), which is a main picture (Key frame) in the video material. Performing a sub-sampling to generate a plurality of sample symbols; and a multiple description coding unit (13) including a plurality of intra-picture encoders (131a) (131b) and commonly connected to the sub-sampling unit (12) to Obtain a quantized main picture composed of a plurality of sampled symbols, and then disassemble them into a plurality of low-resolution pictures, which are encoded by the corresponding intra-frame encoder (131a) (131b), and then belong to the associated channel (CH1) (CH2). External transmission, where low resolution pictures transmitted by the in-picture encoder (131a) (131b) channel (CH1) (CH2) must have similar but independent statistical characteristics.

The inter-picture decoding module (20) further includes: a non-main picture decoding module (21) coupled to the non-home picture coding unit (11) to decode the non-home picture; and a multiple description decoding unit (22) comprising a plurality of in-picture decoders (221a) (221b) coupled to corresponding intra-frame encoders (131a) (131b) to restore low-resolution pictures of corresponding channels; and a super The resolution restoration reconstruction unit (23) is connected to the complex intra-picture decoder (221a) (221b) of the multiple description decoding unit (22), and the complex low-resolution picture combination is restored to the original home picture.

The above-mentioned sub-sampling unit (12) may adopt spatial domain sub-sampling (Spatial Domain) or frequency domain sub-sampling (DCT Domain). If the battery power and computing power of the wireless distributed video codec system application are considered, the spatial domain is adopted. Subsampling is preferred. The space domain subsampling is further described below. Spatial sub-sampling can be further divided into two categories, one is the direct sub-sample method, and the other is the weighted sub-sampling method. The direct sub-sampling method is used to sample the chroma of the picture. The weighted subsampling method is used for the picture brightness sampling portion. The following is explained by the direct subsampling method.

If you want to sample each picture as a quarter dimension, the direct subsampling method divides the video into multiple 4x4 blocks, such as direct left upper sampling and lower right sampling, and thus the second sampling. The picture will form two similar but independent picture blocks to match the encoder channel characteristics in the above picture. Moreover, the direct sub-sampling block is divided into the following four types:

(a) Direct Upper Left (DUL);

(b) Direct Lower Right (DLR);

(c) Direct Lower Left (DLL); and

(d) Direct Upper Right (DUR).

Suppose the pixel of a single main screen is f(x, y) and the dimension is The four blocks of the above direct subsampling can be described as:

Please refer to the third figure. The weighted sub-sampling method is further introduced, that is, the single-sided surface pixel is f(x, y) and the dimension is For example, if you want to divide it into four-picture blocks, take the three-point average method to match the angle between the oblique line and the horizontal line of 0 degrees, and the upper and lower positions of the non-overlapping 2x2 block to divide into the following four areas. Piece.

(a) A3_45_Upper;

(b) A3_45_Lower;

(c) A3_135_Upper; and

(d) A3_135_Lower.

In addition, the pixel points of the three-point average sampling are set to different weighting proportions, and the setting principle is that the ratio of the right-angle pixel to the other two points is 2:1:1. The above four blocks can be respectively described as follows:

g _{A 3_45_ UP ( x , y )} = f (2 x , 2 y +1) + f (2 x +1,2 y )+ f (2 x ,2 y )/3

g _{A 3_45_ LOW ( x , y )} =f (2 x ,2 y +1)+ f (2 x +1,2 y )+ f (2 x +1,2 y +1)/3

g _{A 3_135_ UP ( x , y )} = f (2 x +1,2 y +1)+ f (2 x ,2 y )+ f (2 x +1+1,2 y )/3

g _{A 3} _ ₁₃₅ _ _{LOW ( x,y )} = f (2 x +1,2 y +1)+ f (2 x ,2 y )+ f (2 x ,2 y +1)/3

The following describes the reduction method used by the inter-picture decoding module (20) to restore the direct sub-sampling method and the weighted sub-weighting method.

Please also refer to the second figure, which first explains the direct reduction method of the direct subsampling method, that is, the sampling point is directly replaced by the corresponding position of the restored picture, for example, the DUL sub-sampling point replaces the 4X4 area of the video to be restored. In the upper left corner of the block, the DUL will describe the DLR with another corresponding sub-sampling. The DLR point replaces the lower right corner of the 4X4 block, and the rest of the direct sub-sampling reduction method is equivalent to this, and the restoration can be completed.

Referring to the fourth figure, the block sampled by the weighted sub-sampling method includes the point information of the non-orthogonal position, the unknown point and the picture dimension, and the position of the known point, so the formula (1)(2) can be restored. The two known points A' and D' of the right angle position point reconstruction screen set to the higher sampling weight are restored by the appropriate interpolation methods (3) to (5) with the information of other weighting points. To ensure that the difference between the interpolated value and the original video pixel value is the smallest, as described below.

From the above formula (5), B + C = 2 ( a ₁ + b ₁ ) - ( A + D ) can be obtained. When the two known points A' and D' of the original picture are calculated by the reduction formula (1), and the known B' and C' unknown points are combined, the known points A and D are used for the nearest neighbor interpolation, and B There is an error in C, and B+C and B' and C' are known. Under the condition of the limit, want a set of B" and C", making , that is, the pixel value difference between B"+C" and B+C is smaller than B'+C'. B' and C', which are obtained by interpolating B+C, are known. B' and C' obtained by interpolation from B'-C' must also be similar to B and C values. Otherwise, the correction method will be It does not make sense, so that B' and C' are similar to B and C respectively. By correcting the (B+C) and (B'-C') solutions, the corrected B and C values are B" and C".

Please refer to the fifth figure, which is a processing flowchart of the wireless distributed video codec system of the present invention, that is, the main picture is first sampled by using four pixels as a unit, and two picture coding units are taken as an example. Perform two sub-samplings on the main picture, so there will be at least two unknown points in the restoration. For the above-mentioned sub-sampling and restoration instructions, the Pixel-wise method can be used according to the known points. Interpolation as a still image, in addition, the interpolated image will be further enhanced by image enhancement. When the encoded low-resolution area picture is transmitted to the inter-picture decoding correction group, the two-picture decoding unit obtains the low-resolution block picture from the two channels separately to perform decoding, and then performs error detection to reduce Transmit the lost opportunity of the packet, and finally complete the high-resolution main picture reconstruction, and perform rate distortion on the main picture compressed in the original dimension without sub-sampling in different sampling, reconstruction methods and multiple descriptions. Comparison of curves.

Referring to the sixth figure, the wireless distributed video codec system of the present invention reconstructs a plurality of main pictures under the condition of GOP=2, thereby evaluating that the effect of the multiple description coding technique on the correctness of the side information is relatively low. In addition, the negative impact on the RD performance of the Wyner-Ziv screen is quite limited.

In summary, the present invention mainly provides a multiple description decoding unit for the main picture coding part of the intra picture coding module, and further divides the quantized main picture obtained by the second quantization into a plurality of low resolution pictures, and then The transmission is respectively performed by the corresponding channel that generates the low-resolution picture; therefore, the multi-description decoding unit can disperse the encoding and transmission of the single main picture, and complete the encoding and transmission of the single main picture at a faster speed, if the application The DISCOVER encoder can effectively improve the encoding speed of the main picture.

(10). . . Internal picture coding module

(11). . . Non-home picture coding unit

(12). . . Subsampling unit

(13). . . Multiple description decoding unit

(131a) (131b). . . In-picture encoder

(20). . . Inter-picture decoding module

(twenty one). . . Non-master picture decoding module

(twenty two). . . Multiple description decoding unit

(221a) (221b). . . In-picture decoder

(twenty three). . . Super-resolution reduction reconstruction unit

(30). . . Code side

(31). . . Quantizer

(32). . . Encoder

(40). . . Decoding end

(41). . . decoder

(42). . . Coefficient reconstruction unit

(50). . . In-picture coding single module

(51). . . Quantizer

(52). . . High speed encoder

(53). . . buffer

(54). . . Intra picture coding unit

(60). . . In-picture decoding module

(61). . . Fast decoder

(62). . . Reconstruction unit

(63). . . Intra picture decoding unit

(64). . . Side decoding information generating unit

The first figure is a block diagram of the wireless distributed video codec system of the present invention.

The second figure is a schematic diagram of the principle of direct subsampling in the present invention.

The third figure is a schematic diagram of the principle of direct subsampling in the present invention.

The fourth figure is a schematic diagram of the principle of using weighted subsampling in the present invention.

Figure 5 is a flow chart of the processing of the wireless distributed video codec system of the present invention.

Figure 6 is a comparison diagram of information accuracy and information accuracy of the side of the wireless distributed video codec system of the present invention.

Figure 7 is a block diagram of a distributed video codec system.

Figure 8: Another block diagram of a decentralized video codec system.

Figure IX: System block diagram of the eighth figure in the field of DCT.

(10). . . Internal picture coding module

(11). . . Non-home picture coding unit

(12). . . Subsampling unit

(13). . . Multiple description decoding unit

(131a) (131b). . . In-picture encoder

(20). . . Inter-picture decoding module

(twenty one). . . Non-master picture decoding module

(twenty two). . . Multiple description decoding unit

(221a) (221b). . . In-picture decoder

(twenty three). . . Super-resolution reduction reconstruction unit

Claims (11)

A high-speed encoded wireless distributed video encoding and decoding system includes an intra-picture coding module and an inter-picture decoding module, wherein: the internal picture coding module includes: a non-main picture coding unit, which is to receive video data. The non-main picture is encoded; the primary sampling unit performs sub-sampling on the main picture in the video data to generate a plurality of sample symbols; and a multiple description coding unit, which includes a plurality of intra-picture encoders, and Connected to the sub-sampling unit to obtain a quantized main picture composed of a plurality of sampled symbols, and then disassembled into a plurality of low-resolution pictures, encoded by a corresponding intra-frame encoder, and then transmitted externally by the associated channel. The low-resolution picture transmitted by the encoder channel in the plurality of pictures must have similar but independent statistical characteristics; and the inter-picture decoding module includes: a non-main picture decoding module connected to the non-main picture coding unit, To decode a non-home picture; a multiple description decoding unit, which includes a plurality of intra-picture decoders, and respectively connected to the pair An intra-picture encoder to restore the low-resolution picture of the corresponding channel; and a super-resolution restoration reconstruction unit connected to the complex intra-picture decoder of the multiple description decoding unit, and the complex low-resolution picture combination is restored to the original Picture. The above-mentioned sub-sampling unit performs spatial domain sub-sampling, as in the high-speed coded wireless distributed video codec system described in claim 1 of the patent application. Subsample size is quantified. In the wireless distributed video codec system of the high speed encoding according to claim 1, the subsampling unit performs frequency domain subsampling for quantization. The high-speed coded wireless distributed video codec system according to claim 2, wherein the spatial domain sub-sampling system comprises a direct sub-sampling method and a weighted sub-sampling method. The wireless distributed video codec system of the high speed encoding according to claim 4, wherein the direct sampling method samples each main picture as a quarter dimension, and the direct subsampling method divides the picture into It is a plurality of 4x4 blocks, and samples the upper left point, the upper right point, the lower left point and the lower right point directly to generate four similar but independent blocks. The high-speed coded wireless distributed video codec system according to claim 5, wherein the pixel of the single main picture is f(x, y) and the dimension is , the four blocks are described as: f (2 x ,2 y )= g _{DUL ( x , y )} f (2 x +1,2 y +1)= g _{DLR ( x , y )} f (2 x +1,2 y )= g _{DLL ( x , y )} f (2 x ,2 y +1)= g _{DUR ( x , y )} g ( x , y ) is a sub-sampling single video with dimensions . For example, in the high-speed coded wireless distributed video codec system described in claim 4, the weighted sub-sampling method is configured to match the angle between the oblique line and the horizontal line at 0 degrees in a three-point averaging manner, and the upper and lower positions of the non-overlapping 2x2 four blocks. Come A single main picture is divided to divide into four blocks, and the pixels of the three-point average sampling are set to different weighting proportions. For example, in the high-speed coded wireless distributed video codec system described in claim 7, the right-angled pixel point of the three-point average sampled pixel and the other two points have a weighted specific gravity of 2:1:1. The high-speed coded wireless distributed video codec system according to claim 8 , wherein the single-image surface pixel is f(x, y) and the dimension is And the four blocks are respectively expressed as: g _{A 3_45_ UP ( x , y )} = f (2 x , 2 y +1) + f (2 x +1, 2 y ) + f (2 x , 2 y )/3 g _{A 3_45_ LOW ( x , y )} = f (2 x ,2 y +1)+ f (2 x +1,2 y )+ f (2 x +1,2 y +1)/3 g _{A 3_135_ UP ( x , y )} = f (2 x +1,2 y +1)+ f (2 x ,2 y )+ f (2 x +1+1,2 y )/3 g _{A 3_135_ LOW ( x , y )} = f (2 x +1,2 y +1)+ f (2 x ,2 y )+ f (2 x ,2 y +1)/3. For example, the high-speed coded wireless distributed video codec system described in claim 6 of the patent scope, the inter-picture decoding module performs a direct sub-sampling method, which directly replaces the corresponding position of the restored picture with the sampling point, and directly restores carry out. For example, in the high-speed encoded wireless distributed video codec system according to claim 9, the inter-picture decoding module performs a weighted sub-sample reduction method, wherein the inter-picture decoding module includes a point having a non-orthogonal position. Information, unknown points and picture dimensions, location of known points, through a reduction formula, set the two known points of the picture to the right angle position of the higher sampling weight, and then use the information of other power points to match The interpolation method is performed to restore.