Study On Adaptive Interpolation Filter In HEVC

With the development of all kinds of digital technology, the digital information hasbeen spread all over the life. The information needed to be transported and stored isbecoming more and more. As a result, many scientists are devoted themselves to studyon compressing the huge amount of the video source data. ISO/IEC and ITU-T haveestablished many different video coding standards. ISO/IEC established a series ofMPEGx standards and ITU-T established a series of H.26x standards.The video applications with high resolution, high sampling bit depth and highfidelity chroma format increase the video source data dramatically. Accordingly,compressing video source data more efficiently becomes very necessary. Therefore, inJanuary2010, ISO/IEC and ITU-T began to develop the next-generation video codingstandard, namely High Efficiency Video Coding (HEVC) standard, also known asH.265, and in February2013this standard was formally published. Its goal is to doublethe compression efficiency of H.264on the basis of H.264. The video coding standardH.265is mainly designed based on the video coding standard H.264. It retains someoriginal technologies, improves a number of related technologies and adds some newtechnologies as well. Taking the compression efficiency into consideration, among allkinds of new coding tools, adaptive interpolation filter is very important.In this thesis, we provide an intensive study on the implementation of adaptiveinterpolation filters. We have studied how two kinds of symmetric interpolation filterscomplete the process of symmetry hypothesis, which are filters with full symmetry andfilters with flexible symmetry. And also, we compare the advantages and disadvantagesof four kinds of adaptive interpolation filters, which are two-dimensional non-separableinterpolation filter, separable interpolation filter, directional interpolation filter andcombined adaptive-fixed interpolation filter. Based on these, an adaptive sub-pixelinterpolation filter algorithm based on the gradient information of the image for lumacomponent is proposed in the thesis.On the basis of traditional adaptive interpolation filters, we add the gradientinformation of the image to two important parts, which are symmetry hypothesis of thefilter coefficients and the process of interpolation calculation, in order to offset theshortcoming of the traditional filter methods. Firstly, we calculate the gradientinformation of the pixel. Secondly, we complete the process of symmetry hypothesis according to the gradient information. There are five kinds of symmetric filters that wecan choose. The number of coefficients needed to be calculated for each filter isdifferent. Lastly, we use the gradient information as the weighted coefficients tocomplete the process of interpolation calculation. In a word, the gradient informationplays a vital role in the whole algorithm.In this thesis, we add the proposed scheme to the software platform of HEVC,choose5video sequences and do experiments for three kinds of structures, which areIPPP, IBBB and Hierarchical-B structure. In the experiments, we observe thereconstructed videos, record the YPSNR and bit rate and draw the rate-distortion curves.Then we use the subject and objective methods to evaluate the performance of theproposed scheme. Compared to the traditional interpolation scheme in HM9.0, theexperimental results show that the adaptive interpolation scheme based on the gradientinformation of the image increases the coding efficiency. The proposed scheme can notonly improve the PSNR, but also reduce the bit rate.