Research On Key Techniques For Error Concealment In Video Communication

With the rapid development of network and communication technology, video communicationhas been more and more important in people's daily life. Although the representation of video isstraightforward, the information which video carries is very huge. It is inevitable to compress thevideo sequence before transmission. For the prevalent video compression standards, techniquesof prediction, transform and entropy encoding are widely used, which makes the compressedbitstream be sensitive to channel impairment. So it is a great challenge for the encoded video totransmit in wireless network or Internet. To guarantee the quality of video transmission, it isnecessary to consider some error control mechanisms which are used to remove or restrain theinfluence induced by transmission error. Among so many kinds of error control schemes, errorconcealment (EC) which is performed in the decoder side, is one of the effective ways.Meanwhile, EC method is irrelevant to encoder and do not add redundancy in encoding, whichmake the related research very popular.This paper focuses on the EC techniques in the decoder. In considering the transmission ofreal-time video, three types of EC method are studied as follows:(1) By investigating typical spatial EC techniques, one content-adaptive spatial EC methodbased on multi-direction interpolation is proposed. For the non-texture macroblock, the recoveryis done by an algorithm of multi-direction interpolation based on edge judgement. This methodstudies the impact on the estimation of edge distribution in the lost block by considering edgedetection region and relevant edge judgement. By calculating boundary pixels more accurately,the implementation of direction interpolation is improved. According to directional entropy ofedge distribution in the lost macroblock, whether it is texture block or not can be detemined. Forthe texture block, bilinear intepolaton is computed additionally and the average results of twomethods are combined to achieve final recovery. The EC exprimental results for video sequencesverify the better performance.(2) About choosing the best Motion Vector (MV) from candidate MV set, there exists twobasic criterions which measure the matching degree between corresponding motion-compensatedblock of candidate MV and the surrounding area of the lost macroblock. By means of the schemeof fuzzy reasoning, two basic criterions are combined together and a novel temporal EC method based on fuzzy reasoning is proposed. In order to overcome the disvantage that fuzzy techniqueis independent of training samples, a process of adaptively defining fuzzy membership functionsis applied by analyzing the extracted sample sets. Simulation results show the better performanceover temporal EC approaches using two typical measures.(3) By analyzing the characteristic of MVs in a local region, we point out the property ofclustering among these local MVs. Thus, a temporal EC method for H.264 based on mean shift isprsented. Based on the assumption, mean shift procedure is excuted to recover the lost MVs ofsub-blocks in H.264. The involved problems of bandwidth estimation and weighted coefficientsetting in mean shift are also investigated by estimating the standard deviation of computationwindow. Experimental results also show the better recovery for the corrupted blocks and lowcomputation cost.(4) On the basis of studying associated kernel density estimation in mean shift, MV recoverymethod based on kernel regression for H.264 is proposed. Taking the MV recovery as data fittingin a local region, parametric regression is substituted by nonparametric method. As improvement,approach of nonparameteric kernel regression only considers the influence of local data throughkernel function for MV recovery, which can achieve more accurate approximation. Simulationresults show the better performance as compared with the basic parametric regression method.