Multi-level Optimization On Motion Vector For Motion Compensation Frame Interpolation

With the rapid development of high-definition digital TV and multimedia systems, frame rate up conversation is mainly used to realize the frame rate conversion between different video sequences. To satisfy high quality visual enjoyment, how to improve frame rate becomes a hot research topic. Various algorithms of non-motion compensation for static scene or a little movement can get a better visual effect, but will appear jerkiness and blurring in a larger movement scene. Motion compensation which takes object motion information into account is another kind of approaches. It includes two processes:motion estimation and motion compensation. But it is still difficult to obtain true motion vectors. Traditional motion estimation algorithms easily result in blocking artifacts and ghost effects for a larger movement too. In this paper, we take motion estimation and motion vector processing as two breakthrough points, and put forward an algorithm based on weighted motion estimation and hierarchical motion vector processing, which effectively improves the accuracy of the motion vector and the quality of the interpolated frame. The main contributions are as follows:1:The paper analyzes defects of the block-matching algorithm based on unidirectional and directional motion estimation, and puts forward a weighted motion estimation method to optimize the motion compensation. The method takes the block boundary correlation and a new constructed measurement criterion into account. It effectively eliminates the blocking artifacts and improves the accuracy of the initial motion vectors.2:Unreliable motion vectors mostly exist in the moving areas. It is necessary to classify motion vectors in each frame. Each motion vector of a block reflects its trajectory, with direction and size. By the characteristic of similar blocking clustering, we adopt an improved k-means algorithm to split each frame into moving and background blocks.3:The paper presents a hierarchical processing method to correct the motion vectors of the moving areas. It can effectively eliminate the blocking artifacts and ghosts. The method is a gradually optimization process which consists of three steps: MV pre-screening, MV reclassification and MV smoothing. The process not only protects the moving areas' edge structure from being damaged by block segmentation, but also gets more detailed information and improves accuracy of the motion vectors.4:Experiments show that the algorithm not only improves the image quality both subjectively and objectively, but it also has a better adaptability to video sequences with fast motion and complex background.5:At last, we compare the algorithm with the different estimation methods, and vector post-processing methods respectively. Experiments verify the effectiveness of the weighted motion estimation and hierarchical processing in our algorithm.