Research On Energy-concentrated Hyperplane Of Multi-dimensional Vector Matrix

There are many ways to express the information. Information can be stored andtransmitted by texts, charts, photos, videos and so on. Nowadays, the greattechnological progress has ensured us more choices to obtain the information besidesthe traditional ways like papers, broadcasting and TV news. With more and moredeveloped Network and Multi-media technology, massive, real time information indigital type has connected every aspect of daily life, which brings great changes in life.Among all the methods of information dissemination, videos are the most vivid andexplicit. The huge amounts of information are the most prominent feature of videos. Totransmit and store video information with higher quality on limited bandwidth, videocompression with high efficiency and reliability is unavoidable and will always be thekey research in this area.The necessity to compress the videos is based on people’s needs, while thefeasibility to compress the videos is based on video pictures’ inherentcharacteristic—the existence of huge amounts of redundant information. To bestcompress these redundant information under certain circumstances is the goal of videocompression. To reach the goal, hundreds and thousands of brilliant scholars at homeand abroad have made their efforts in probing, and a series of effective and reliablestandards and algorithms are put forward. The most widely used standards are MPEGseries and H.26x series. The high efficiency video compression standard is the newest ofall the standards up to now, which is released in the year2013. There are severalprocedures in the whole video compression process, among those the most importantand time-consuming procedure is motion estimation. The performance of motionestimation algorithm determines the performance of the video compression. Likewise,the speed of motion estimation directly determines if one certain algorithm is suitablefor real-time transmission. Fast motion estimation algorithms are brought up to meetthese needs, among which HEXBS is widely used. The line fitting algorithm proposedin this paper is based on the algorithms mentioned above, for the purpose of betterperformance and higher accuracy.The algorithm presented in this paper is based on the MVM (Multi-dimensionalvector matrix) theory proposed by my lab. In theoretical part, the conceptions of MVMand matrix partition method are introduced. Some theoretical derivations are displayedwhen MVM theory is applied to DFT domain and DCT domain. The block-diagrams ofsome existing motion estimation algorithms are given briefly. Analysis and conclusionsare made on their characteristics and applicable scope respectively. Artificial video sequences are chosen to realize this algorithm in the experimental part. A3Denergy-concentrated hyperplane can be obtained after the artificial video sequences aretransformed in frequency domain. This hyperplane contains the direction information ofmotion estimation. The algorithm is focused on how to extract the implicit motioninformation from the hyperplane and give a continuous precision result of motion vectorestimation.Visual C++and MATLAB are used to perform the experiment. In experimentalpart, multi-dimensional video partition and recombination is firstly done. A series offixed size blocks are applied to MVM-DCT in frequency domain after that. Then the3Dfolding plane is obtained through fitting the coefficients after MVM-DCT. The foldingsurface can be extracted and flank information can be acquired through the least squareline fitting method. Finally the motion vector is obtained. The detail chart of theproposed algorithm is given in this paper; the feasibility of the algorithm is alsodiscussed. Compared to plane fitting algorithm, less data is applied during the fittingprocess, which means great fewer burdens in computation. Furthermore, the method ismore accurate in extracting the dot information of the energy concentration plane. Theadaptive window is innovatively adopted in the algorithm. The performance isobviously better than that of the fixed window, with faster convergence speed andacceptable accuracy. The highest precision of the existing video compression standardsis one fourthpixel. Compared to that data, the precision of the algorithm proposed in thepaper is improved significantly, which can reach up to the order of10-4.