Research On Low-complexity Video Codec Using Distributed Arithmetic Coding

Distributed Video Coding(DVC)is different from traditional video coding,which can shift the modules with complex calculations such as motion estimation and motion compensation from encoder to decoder.Therefore,it can be applied to scenarios where the computing power,storage space,and power consumption resources of encoder are limited.Nowadays,most DVC schemes adopt channel coding to implement the source codec,such as Low-Density Parity-Check(LDPC).However,LDPC performs poorly on short to medium block length sources that are commonly seen in video.In addition,LDPC relies on the sparse check matrix during encoding and decoding,which makes its codec structure too complicated.This leads to a problem that DVC encoding performance is suppressed.At the same time,although non-feedback DVC reduces the network delay because there is no feedback channel,it requires higher accuracy of bit rate estimation.In addition,when the bit rate is limited,the prediction of frame is often not accurate enough,which will directly affect the decoding performance of the DVC system.In order to solve the above problems,the following research is carried out in this thesis,which is dedicated to improving the coding efficiency of DVC,improving the rate distortion performance,improving the rate control algorithm,and reducing the complexity of the codec.1)Aiming at the problem that the LDPC-based DVC has a poor decoding performance for short to medium block length sources,this thesis proposes a DVC scheme based on Distributed Arithmetic Coding(DAC).According to our research,it is found that the source coding such as DAC can also be used for related source compression,especially for short to medium block length sources,its compression performance is better than LDPC.Therefore,this thesis is based on the theory of distributed source coding and uses DAC to construct the DVC codec,so as to solve the problem that DVC coding performance is suppressed.2)Aiming at the problem of non-feedback DVC bit rate control,this thesis establishes a bit rate estimation model based on the above DVC codec.In this thesis,a large number of DAC interval overlap factors are mapped to the source entropy.According to the mathematical relationship between source entropy and bit rate,the relationship between the bit rate and the interval expansion factor is found.Then this thesis derives three functional relationship models,and experiment to choose the appropriate functional model.Finally,this thesis uses this function model to implement the rate control of DVC to adapt to complex and changeable video streams.3)Aiming at the problem that the prediction of frame is not accurate when the compression rate is limited,this thesis proposes a frame prediction based on neural network.Based on the outstanding ability of neural network in image analysis and graphics processing,this thesis applies neural network to DVC and propose frame prediction based on neural network.In this thesis,encoding network module,binary network module and decoding network module are used to construct our neural network-based DVC model with a hierarchical network structure.The prediction of key frame and reference frame is realized under the condition of low compression rate.In view of the shortcomings of DVC based on channel coding,this thesis proposes a low-complexity video codec based on DAC,which can reduce the algorithm complexity of DVC codec and improve the compression performance of DVC system.The experimental results show that DAC-based DVC is better than DVC system using channel coding,which can effectively improve the performance of codec.For videos with different intensity of motion,the bit rate estimation model in this thesis can adaptively calculate the appropriate DAC interval overlap factor and achieve code rate control without feedback DVC.At the same time,the encoding complexity of DVC in this thesis is less than 19～25 times of that of decoding,so it can be applied to scenarios with limited resources on the encoder.In addition,under the condition of low compression rate,the neural network can efficiently predict the target frame,and has a good application prospect in the field of DVC.