| With the fast development of multimedia technologies, video compression is the very important motivity for all the exciting and new-style video products. Many areas need more efficient and real-time compression technologies for digital image processing. Fast and efficient compression algorithm is the key for this problem. Therefore, encoding and decoding technologies are continuously improving in the past ten years. Coding techinology itself and the application backgroud of this domain have been changed remarkablely. As a result, there are more and more new promising stars. At present, four source standards can be chosen at home and abroad: MPEG-2,MPEG-4,AVC and AVS. Classified by constitutor, the first three ones are constituted by MPEG expert group and the fourth is by our country; Classified by developing phases, MPEG is the first generation source standard and the others belong to the second generation. Comparing by main technology factor:coding efficiency, MPEG-4 is 1.4 higher than MPEG-2; AVS is almost the same with AVC and they are more than twice comparing with MPEG-2.AVS is the second generation source coding standard which China owns independent intellectual property. Nowadays, the number of AVS team members has reached 162, among them, 25% are Chinese colleges and institutions, and 30% belong to the foreign companies and institution. The industry league covers the whole industry chain, including front end equipments, system provider, TICchip provider, STB manufactures, and relative industry is under developing. On the application prospect of AVS, IPTV developed by the CNC and CMMB for the mobile TV has been compatible with AVS, and more applications can be seen in the fields of terrestrial TV broadcasting and high density CD. It's believed that in the near future, AVS standard will have great prospects. As a result, the research for the key algorithm in AVS is quite necessary.In inter-frame prediction coding, fractional pixel motion compensation is applied in order to achieve more accurate motion vector and higher compression efficiency, and it is the key part in digital video technology. Motion estimation and compensation with half-pixel or quarter-pixel accuracy is supported byAVS, which can improve the accuracy of motion vector. It should be possible to reduce the inter-frame residual error and improve compression performance. However, fractional pixel interpolation algorithm is very complex. It can bring remarkably calculation and increase the complexity in motion estimation process.It can be seen by practical testing that many modules such as DCT, IDCT, motion compensation, interpolation take a lot of time. So it is necessary to optimize these modules in order to improve the coding efficency.This thesis is mainly based on the SD real-time encoder project developed by center research academy of SVA group in Shanghai, with the introduction of fractional pixel interpolation in inter-frame coding based on AVS standard, This part is optimized and a feasible optimized algorithm which can be used in practice is provided.This paper introduces the sub-pixel interpolation theory of AVS and gives a feasible algorithm in practical application.Firstly, this thesis introduced the research background of video coding technology and analyzes the in and abroad video coding standards, gives out the meanings of research.Secondly, based on the inter-frame coding technology based on AVS standard, AVS1-P2 coding structure is described and the inter-frame coding technology is introduced in detail.Then, this thesis optimizes the algorithm of fractional pixel interpolation, and the main process is: in motion estimation, firstly we use UMHexagonS fast full pixel motion search algorithm, then use CBFPS fast fractional pixel motion search altorithm. If the searched motion vector is in the full pixel position, we can fulfill the motion compensation and then inter-frame prediction encoding according to the relative full pixel position of referenced frame; If it is in the fractional pixel position, use the following method: according to the value of motion vector, choose the sub-pixel accuracy and then interpolates referenced sample, that is, make sure the sub-pixel interpolation position based on the values of abscissa and y-axis. There are four cases for mv_x &3: 0, 1, 2, 3, respectively standing for horizontal full pixel, a quarter-pixel, half-pixel, three quarter-pixel interpolation; The same, mv _ y &3 has 0, 1, 2, 3 four cases, respectively standing for vertical full pixel, a quarter-pixel, half pixel, three quarter-pixel interpolation. So taking both mv_x &3 and mv _ y &3 into account, there are 16 cases, corresponding to 16 different pixel interpolation positions. Supposing advalue = (m v_x &3<<2)+ mv_y &3,then the above 16 cases can be expressed by by advalue . We can figure out these 16 cases by full pixel and then store them in the data blocks. When the searched motioh vector is in sub-pixel position, we use one certain case stored in the data blocks beforehand judged by advalue , so there is no need to interpolate on field. We get high visiting speed but cost large memories. Therefore the speed of motion compensation can be improved, what's more, the encoding efficiency can also be increased.Meanwhile, we optimize the programme for remarkable calculating datas in the process of pixel interpolation by the mixed injunctions of C programme and compilation programme, especially using MMX injunctions, as these injunctions can be used in VC++6.0 directly.Finally, the relative simulation results about the proposed algorithm are given in this thesis. By using two video sequences as example to encode, comparing analysis are made for this real-time encoder with standard AVS encoder in the quality of video image, encoding bit rate and encoding speed and conclutions can be drawn as follows: the proposed algorithm is practical, less complex and less impacts on the video quality. This fractional pixel interpolation algorithm can achieve both in high efficiency and accuracy, so it has great practical value in AVS codec. |
PDF Full Text Request