Motion Vector Estimation In Multi-dimension Vector Matrix-DCT Domain

With the continuous development of network information technology,multimedia communication on technology and digital technology, the explosivegrowth of information bring a new experience and great convenience to people’s lives,work and entertainment. One of the effective ways for people to obtain informationfrom outside is rich video information, with the deepening of the digitization processand the information technology, the vivid, fast and efficient for video informationdemand is also more and more high, the video images are spread more widely, so theamount of information transmitted is also growing. Such a huge amount of data thatmakes video image transmission and storage is very difficult, so the video imagecompression effectively has become one of research hotspots in the video field.Motion estimation technology has always played a very important role in videocompression encoding and it is the core module in the video compression. It not onlycan play a role in reducing the redundancy of video signals, and can greatly improvethe coding efficiency. Whether the past H.264(AVC) and the newly establishedstandard H.265(HEVC), high quality of motion estimation is always the basis andprerequisite for efficient video compression, it accounts for about70%of the totalvideo coding calculation above and determines the effect of video compression. Solooking for a fast motion estimation algorithm has become the focus in the field ofvideo coding. This paper describes the principle of the block-matching motionestimation algorithm and several classic block matching motion estimation algorithmsincluding full search method and some fast search method are introduced. Then weanalyzed the principles and processes of these algorithms. For the features that theyare both in the time domain, the method of grid computing, the amount of calculationis large and the search precision is discreted, this paper proposes a new motionestimation algorithm--motion vector estimation in MVM-DCT domain.Our lab proposed the multi-dimensional vector matrix theory and the MVMorthogonal transformation technology is introduced into the theory. Based on theMVM-DCT and using the frequency-domain motion estimation algorithm, it caneffectively eliminate the space redundancy and the time redundancy of video imageinformation.According to the spectral characteristics of the artificial video sequencegenerated by the global constant translation motion in the transform domain can be used for the motion estimation method of the video sequence, this article introduces itto the system of multi-dimensional vector matrix and deduced that video sequenceafter transformation would generate special energy footprint in the frequency domain.The spectrum is limited in one plane and the plane indicates the direction of motion.So this paper proposes a new motion estimation algorithm--motion vector estimationin MVM-DCT domain. First, get the artificial video sequence of a frame of image,which is through the global constant translation motion, then transform it in the DCTof multi-dimensional vector matrix and get it transform coefficients. Then we find thespectrum of plane by using iterative fitting method through the transform coefficient.Finally we can find the corresponding motion vector.In the experiments, this paper get through different directions of iterative fittingfor the folding surface of the same motion vector and verify the validity andfeasibility of this method. It does not need to block matching, can greatly reduce thecomplexity of the calculation. In the MVM-DCT domain, motion estimation accuracyis continuous, so the accuracy error can be reduced to the magnitude of10-2, themotion estimation accuracy can be greatly improved compared to the H.265standardquarter.