Research On Fast Algorithm Of Video Coding

With the rapid growth of computer technology and network,the main carrier of information dissemination has been gradually replaced by video,and video data volume has also become the main body of network big data.With the emergence of high-definition and ultra-high-definition videos,the space occupied by videos has gradually increased.In order to efficiently compress and store data and transmit it quickly on the Internet,video encoding technology has become the key.Therefore,the Joint Collaborative Team on Video Coding(JCT-VC)has released new generation of High Efficiency Video Coding(HEVC)standard.Compared with the previous generation of Advanced Video Coding(AVC)standard,HEVC standard saves about 50% of the code rate under the same visual quality,but because of the introduction of extremely complex coding technology,its coding complexity increases sharply.Therefore,for the better development of video coding technology,it is very important to study the fast algorithm of video coding.In order to solve the problem of high complexity,this thesis adopts the outstanding feature extraction ability and excellent learning ability of convolutional neural network,and designs a fast partition algorithm of intra-frame prediction Coding Unit(CU).In video coding,intra-frame prediction coding is one of the key technologies,and its main purpose is to remove video spatial redundancy.After research and analysis,it is found that the computational complexity of the CU partition decision process in intra-frame predictive coding is very high.In this regard,this thesis proposes a fast partition algorithm of intra CU based on a deep learning.The algorithm first designs a two-level neural network model.The first level uses the branch structure to extract the low-level features of the image,the second level uses the Inception module to extract the high-level features,and then uses the HEVC official test model HM16.20 to obtain data set of the CU and its corresponding partition.The set is used for the training and learning of the model,and finally the trained model is integrated in HM16.20,so that the CU partition decision is obtained from the model,thereby skipping the complex rate-distortion optimization process.The experimental results show that,compared with the original HM16.20 encoder,the encoding time of this algorithm is reduced by 60.52% without the loss of encoding performance,and the encoding complexity of HEVC is reduced.For the above model,there is a problem that the partition of Coding Tree Unit(CTU)cannot be predicted at one time,this thesis proposes an efficient partition prediction algorithm of intra CU based on deep learning.The algorithm first designs an efficient two-level neural network model.In order that the output of the model is a CTU partition decision,the CTU partition is efficiently modeled.The first level of the model uses a three-branch structure to extract low-level image features.The second level still uses the Inception module to extract high-level features,and then is trained and learned through the data set,so that the model can predict the CTU partition decision at one time,thereby skipping a lot of computational redundancy.The experiments show that the coding time of this algorithm is reduced by 64.98%with little loss of coding performance.