H.264 / AVC Intra Prediction Mode Quick Selection And Motion Estimation Algorithm

As a new video compression coding standard, H.264/AVC increases many new technologieson the basis of keeping former advanced technologies. Such as: multiple reference frame, ratedistortion optimization technique, integer transformation, looping iflter, new entropy coding andso on. Though the introduction of new technologies make H.264/AVC embody more advantages,it also complicates the design of the codec，limits the scope of application ifelds and so on. Sothrough improving coding algorithm to reduce the computation complexity is very important.This paper ifrst introduces the research signiifcance and backgrounds overview of videocompression coding standards. Though introducing the development history of the codingstandards we educe H.264/AVC. Then it elaborates on the theory, the structure and the relativetechnical supports of the new standard’s coders and decoders. Intra-frame prediction searches thebest prediction model through traversing all possible modes，so the algorithm is very complicated.We propose a fast method on prediction mode selection in this paper, and it can improve thecoding eiffciency. While wavelet domain motion estimation overcomes block effects and largecomputation in spatial domain motion estimation, but there is shift-variant. In the motionestimation process, it increases the dual-tree complex wavelet decomposition process and the ideaof hierarchical motion estimation to reduce the shift-variant of wavelet transformation and thecomplexity of motion-estimation. Concrete research contents as follows:(1)Determination before intra-prediction. Through calculating SSD values of macro blocksto determine whether residual blocks are all-zero blocks or not. Threshold values are set as T andresidual blocks are determined as all-zero blocks if SSD values are less than T. The methodavoids all-zero blocks’ process of transformation, quantization, inverse transformation andinverse quantization and then improves the speed of coding.(2)Fast intra-prediction mode selection algorithm for H.264based on texture. The time thatthe luminance4x4mode expands takes a great percentage in the whole intra prediction process,so we put forward a modiifed4x4luminance blocks mode choosing method. In the case thatresidual blocks are not all-zero blocks，Canny edge detector calculates the gradient direction ofall pixels in the image, combines with square gradient method to calculate the image’s texturedirection and then chooses the corresponding probable modes in the luminance4x4predetermination mode on the basis of the texture direction. This new method not only saves theprocess of handling all-zero blocks but also lessens the scope of optimum mode searching,simpliifes the amount of computations and accelerates the speed of coding.(3)Research on intra-prediction mode selection algorithm for H.264/AVC. The paper putsforward a method to use the gradient magnitude to determine the degree of lfatness of macro blocks. Kirsch edge detector calculates the gradient magnitude and the gradient direction ofcurrent macro blocks. The gradient magnitude is used to predetermine that the choice of mode is4x4or16x16and the4x4mode is directly chose for the image with relative details. Moreover,the gradient direction is used to determine the choice of luminance4x4and16x16predetermination mode.(4) Research and application on high precision in motion estimation based on dual-treecomplex wavelets. The paper puts forward a kind of improved dual-tree complex wavelettransformation method. The main idea of the method is to improve the translational variability ofwavelet-domain motion estimation. After using the proposed method to conduct waveletdecomposition with the current frame and the reference frame, it conducts hierarchical motionestimation with the sub-bands. The new method enhances the accuracy of wavelet-domainmotion estimation because of avoiding directly dealing with high frequency information withhigher sensitivity.We take JM13.2, the reference model of H.264/AVC as the test platform, aiming at differentvideo sequences, it conducts simulation of the proposed algorithm. The proposed algorithmlessens the complexity of calculation, reduces the time of coding and improves the searchingaccuracy in the process of motion estimation under the condition that the peak signal ratio and thecoding rate keep essential balance.