Research On Frame Rate Up-Conversion Algorithm

Frame rate up-conversion can convert low frame rate video into highframe rate video through insert interpolated frames into original framesequence. This technology was proposed in1980s, which has been widelyapplied with the development of digital TV. In low bit rate video transmissionarea, enhance video frame rate at receive end can overcome the influence ofskipped frames, smooth object motion, improve visual quality. While inmultimedia video broadcasting field, this technology can effectively reducemotion blur result from hold-type rendition of LCDs.Frame rate up-conversion algorithm mainly consists of two processes:motion estimation and motion compensation. Motion estimation is a processthat computes the motion vectors of two neighborhood frames. For its lowcomputation and easy to implement in hardware, Block Matching Algorithm(BMA) is the most commonly used algorithm. However, the BMA alwaysselect block sum of absolute differences as unique criterion to evaluate thecorrectness of motion vectors and cannot accurate estimate motion vectorsin motion object boundary, which influence the final interpolated framevisual quality. Motion compensation adopts motion vectors achieved frommotion estimation to construct interpolated frames. It can divide into twocategories: motion trajectory interpolation and dual motion compensation.Dual motion compensation always adopts overlapped block motioncompensation method to reduce block artifacts, but can cause blurphenomenon in boundary areas.For the boundary motion vectors distortion and overlapped block motion vectors selected unaccuratly problem which existed in block basedmotion estimation, we combined motion segmentation algorithm and depthorder acquisition algorithm, proposed a modified motion estimation method.This method first adopts pyramid hierarchical search algorithm to acquirerelatively accurate unidirectional motion vectors. Secondly, we segmentedthe motion field and used depth order acquisition algorithm to compute thedepth order of each motion regions. Then, we considered minimum sum ofabsolute differences and depth order information to convert unidirectionalmotion vectors into bidirectional motion vectors. At last, we decomposedblocks, smooth and correct motion vectors. Experimental results show thatour method can enhance smoothness and correctness of motion vectors.For the motion boundary region blur and interpolated frame mergecomplex or uneffective problem existed in overlapped block motioncompensation, this paper proposed a modified adaptive overlapped blockmotion compensation method based on motion segmentation optimization.Firstly, we used the similarity of the neighborhood blocks to determine thepre/back frame weighting coefficients. Secondly, we judged the size anddirection of one expansion overlapped block through the motionsegmentation result. Finally, we adopted the mean SAD acquired in motionestimation to do interpolated frames adaptive merge, which consideredeffective and efficient at the same time. Experiments prove that our methodcan alleviate the boundary blur and distortion phenomenan.Finally, we modified convertional frame rate up-conversion algorithmbased on our motion estimation method and motion compensation method.Then, the paper compared our proposed algorithm with two other classicalalgorithms and find that our algorithm has better effects both subjectively and objecti vel y.