Research On Side Information Generation Of Distributed Video Coding

With the rapid development of information technology and wireless network, wirelessmobile terminal is becoming an important way for multimedia communication. Theapplication of mobile terminal equipment is limited by device’s computing capability andpower consumption. However, traditional video coding standard has the characteristics ofhigh computational complexity in encoding while low in decoding,which is not suitable formobile terminal. Distributed video coding (DVC) allows for shifting the computingcomplexity from encoder to the decoder, and also has a high video transmission robustness, soit is very suitable for mobile video applications. As one of the key technologies for distributedvideo coding, the quality of Side Information (SI) directly affects the performance of DVCsystem. This paper will focus on SI generation algorithm for DVC. As the estimation ofcurrent WZ frame to be decoded, the more accurate SI is, the less check bits decoder needsand the better performance rate-distortion has. This paper puts forward two kinds of SIgeneration algorithms based on current SI research. The main contents include:(1) This paper discusses the basic principles of DVC, the DVC system framework,DISCOVER codec framework, and studies the typical algorithm for the generation of SI indetails.(2) Having studied SI generation using motion compensated temporal interpolation(MCTI), this paper presents a kind of SI generation algorithm based on Hash code, which isdependent on the broken-line motion estimation interpolation approach. Considering themacro-block regions abound that do not meet the assumption of SMV in real videosequences,the algorithm analyzes the motion features of video sequences and gradually putsforward some new motion models: linear motion in a straight line, broken-line and curvemotion. According to the correlation of macro-block in the WZ frame and Key frame, itcorrespondingly transmits the hash information to decoder. Using the hash information, thealgorithm extracts motion vector based on asymmetrical polygonal model, and then moreaccurate side information is obtained. The experimental results show that the proposedalgorithm is valid.(3)According to macro-block curve motion assumption, a new SI generation algorithm using curve motion estimation of block segmentation is proposed. The new algorithm makes adeeper segmentation on the macro-blocks with intense motion, then redoes curve motionassumption in successive Key frames. Next, after having acquired coordinates of similarmacro-blocks，the algorithm will determine the curve trajectory by linear regression. Finally,SI is generated with different weighted values. Experimental results show that the proposedside information extraction method enhances the accuracy of SI effectively compared to theMCTI algorithm.