Research On Energy Concentration Analysis Of Multi-view Video Based On Multi-dimensional Vector Matrix Orthogonal Transformation

With the development of science and technology, the requirement for multimediacommunication has expanded from the traditional flat-screen video to multi-viewvideo. The so-called multi-view video refers to a series of video shooting obtained tothe same scene, on the different angles at the same time. For example: the commonmovie video is a view of the video, also known as single-view video; the popular3Dmovie is the two view of the video.As we all know, a singal-view movie video will take up hundreds of Mbit orseveral Gbits memory space, and a multi-view video equals N single view videos, sothe amount of data it contains is N times that of the corresponding single-view video.The huge amount of data generated by the multi-view video will be a great impact onstorage of the source and transmission of the channel. Taking into account a certainangle between multiple cameras of the multi-view video, the multi-view videosequence will contain a certain amount of duplicate information. Thereby thisgenerates the inter-view redundancy. While the video contains a large number ofinterframe redundancy and spatial redundancy in the internal, thus they make themulti-view video compression coding technology be achieved.The multi-dimensional vector matrix theory extends the traditionaltwo-dimensional matrix to the multi-dimensional level. Since the multi-dimensionalvector matrix can put the multi-dimensional data of multi-video video into amulti-dimensional model, so it can remove completely redundant among eachdimension, and it also replaces the motion estimation. It uses two vectors to representthe multi-dimensional orthogonal transform operators and unifies multi-dimensionalmatrix algorithms. In the process of computing, it first uses the matrix to express eachoperator specifically in order to invoke facilitately, so it can save a lot of computingtime.In this paper, we based on multi-dimensional vector matrix theory, firstestablished in4D mathematical model of multi-view video, then combined the Fourier transform (FT) and multi-dimensional vector matrix discrete cosine transform(MVM-DCT) respectively with multi-view video model established, and derived aorthogonal transformation formula under the significance of multi-view video.Integrated all the advantages both multi-dimensional vector matrix and discrete cosinetransform, the MVM-DCT derived can completely remove the redundant amongspace, inter-frame and inter-view.The deduced formula shows that coefficient energyin high-dimensional transformation domain is mainly concentrated in a folding cuboidwhich attenuates along both the time dimension and view dimension. The significanceof formula derivation is that they reveal the relationship between motion vector, anglevariable and coefficients concentration in high-dimensional transformation domain.Subsequently, according to the coefficients energy distribution law derived ofmulti-view video after the orthogonal transform, the two multi-view video adaptivescanning algorithms are proposed, i.e. the multi-view video adaptive scan by size(MVASS) and multi-view video adaptive scan by position (MVASP). MVASS sortedthe transformed coefficients from big to small, and retained some front biggercoefficients according to the requirements, set the rear smaller coefficients to zero,and at the same time recorded the positions of the non-zero coefficient, in order tocarry out decoding. This method can obtain optimal energy packing efficiency (EPE).MVASP is based on the plane of coefficient energy concentration, retaines the nearplane coefficients as much as possible, and set the far plane coefficients to zero. In thecase of the known motion vector, the method needn’t record the positions of the nearplane coefficients. Therefore, compared with MVASS, it can save a lot of bit rateoverhead.Finally, the simulation experiments are operated under the Windows XP system,on Visual C++6.0platform. Both simulation result and objective data prove thecorrectness of the theory derivated of this article. Comparated with traditional Zig-zagscanning algorithm, it also fully embodies the superiority of the two scanningalgorithms proposed in this paper and proves further the correctness of theoretical derivation about energy concentration analysis.