Research On Multi-view Video Coding Based On Multi-dimensional Vector Matrix

With the rapid development of the information technology,multimediacommunication is increasingly becoming an indispensable part of humanlife.Especially in recent years,the development of3D stereoscopic video has drawnmore attention.From the3D stereoscopic movies in cinema to the3DTV for homeuse,stereoscopic video undoubtedly brings a brand-new,more realistic visualexperience.Common3D stereoscopic video,is to simulate the human eyes,by two camerasshooting the same object from different angles at the same time.At the playbackstage,is to separately giving the pictures shooting by the two cameras to botheyes,forming a stereoscopic feeling in mind.In order to achieve a stronger sense ofthat feeling,sometimes we use more cameras.That’s what we call the multi-viewvideo.Talking about the development of video technology,it is inevitably linkedtogether with the video coding.Because it usually contains enormous data,and hasbecome a great barrel in video communication and transmission.Traditional videocoding techniques have come to a muture stage,their research has reached theperformance bottleneck.However,the research of multi-view video coding(MVC) isstill on the first stage,many theories need to be improved.Therefore,our lab proposed the innovative multi-dimensional vector matrixtheory.It breaks the limitation of the traditional2-dimensional model and promotes theconcept of matrix to multi-dimensions.Based on this,we proposed themulti-dimensional vector DCT orthogonal transform kernel matrix based on discretecosine transform(DCT),and verified its orthogonality.In the previous studies,werealized the coding of color video stream based on this theory.Consider the use of thismulti-dimensional model in the processing of multi-view video can effectivelyremove the strong correlation among each angle,so,the purpose of this article,is to realize the multi-view video coding based on the multi-dimensional vector matrixtheory,and has already made some achievements.In this article,first,we did the block processing of the multi-view video sourcedata,in order to do the DCT orthogonal transform later;Consider the correlationamong each angle,we reorganized the blocked data according to each angle’sorder.The reorganized data constitute the original data involved in the DCT transform.According to the mathematical model of the multi-dimensional vector DCTorthogonal transform kernel matrix,we did the multi-dimensional vector DCTorthogonal transform to the reorganized data,the transformed coefficients had a goodeffect of concentration.According to the distribution of the coefficient,we used a moreflexible multi-dimensional quantization method,to remove some information uselessto the video.The Final work are the differential encoding,multi-dimensionalscanning,and run length encoding.The decoding process is the reverse process of thewhole coding process.At last we achieved the programming of the scheme in this article in VC++environment,discussed the compression performances in different quantizationconditions,verified the effectiveness and the feasibility of the scheme in thisarticle.Research on the multi-view video coding based on the multi-dimensionalvector matrix theory is still in its infancy,many steps in this scheme also have largeroom for improvement,we hope that this research may lay a good foundation forfurther study.