Research On MERGE Mode And Related Technologies Of The Next Generation Video Coding Standard

Motion estimation and its corresponding motion compensation is the core partof the video coding standard. As a result, the encoder just need to transmit themotion information and residual information to the decoder without transmitting theoriginal pixel values, which greatly improves the compression efficiency. However,passing the motion information to the decoding side needs a considerable part of theencoded bit stream, especially in low bit rate case. Therefore, how to encode themotion information is the focus of our research.Skip, direct and the emerging merge mode of the new generation of videocoding standard High Efficiency Video Coding（HEVC）can effectively describe themovement of a block without having to pass any additional motion information. Allof them predict the motion information of the current block from the motioninformation of neighboring blocks. Merge mode has been widely used in HEVC andHEVC-3DV currently. This paper mainly studies how to improve the predictiveeffect of the merge mode.This paper first proposes a program to optimize the prediction candidate set ofmerge mode. Motion vector prediction candidate set of merge mode contains themotion information of the left, top, top right, bottom left, top left and bottom rightneighboring blocks. Motion information of the below and right adjacent blocks isnot used effectively. This paper adds their motion information to the predictioncandidate set of the merge mode to ensure that the motion information of currentblock can be predicted in all directions. The program has better performancecompared to HM4.0.Both the new generation of video coding standard HEVC and the current videocoding standard H.264/MPEG-4AVC employ block matching motion estimation(BMME) which is based on translation motion model. This makes it difficult torepresent the complex motion accurately such as rotation, zoom, and etc. In thispaper, we propose an adjacent-block-based prediction model to improve theprediction performance of a deformable block. We integrate our method into HEVCas an additional choice of its merge mode. Based on this model, the motioninformation of each4x4block of current prediction unit can be derived from themotion information of the nearest neighbors to its four corners. Simulation resultsshow that our proposed method has better performance compared to HM4.0, the BDbit rate saving is up to15.4%, while the encoding and decoding complexities arealmost the same.Although temporal prediction is usually more effective than inter-view prediction, they have their own advantages. In some cases(such as for fast-movingobjects), inter-view prediction has better prediction performance. Generally, thetemporal motion information and the inter-view motion information don’t have anycorrelation. However, the merge mode of HEVC-3DV doesn’t distinguish temporalprediction candidate and inter-view prediction candidate. To improve the predictiveeffect of the merge mode, we distinguish the prediction candidates in temporal andview direction. Accordingly, there are two kinds of merge mode: temporal mergemode and inter-view merge mode. The experimental results show that the programhas better coding performance than HEVC-3DV-0.4in general without increasingcomplexity at both encoder and decoder.