The Video Error Concealment Using Inpainting Methods

With the development of network and video compression technology, the application of transporting video signal through wireless channel and Internet has been more and more common. The compressed video signal will be susceptible to the channel disturbance, and cause error, resulting decline in the quality of video. Error concealment is an important technology to improve the video quality on the decoder, without increasing the transmission channel bandwidth, or the amount of information.According to repair theory, error concealment can be divided into spatial domain and time domain two categories, each with different applications. Image inpainting is an image restortion technology based on the information of the avalible point around the damged area. Image inpainting can be broadly divied into two types: structure reconstruction and texture synthesis. Video inpainting technology aimed to inpaint image sequence, including temporal inpainting and motion inpainting.In this thesis, we first study spatial and temporal error concealment methods in H.264, and introduce image inpainting and video inpainting methods, and then proposed improved algorithm for both spatial and temporal error concealment. For spatial error concealment, a new method based on the integration of Exemplar-Based Image Inpainting and FMM-based Image Inpainting is proposed for the macroblock lost condition.We will process the lost macroblock in 4x4 block unit. After calculating the priority of each 4x4 block in lost macroblock, we will find the best matching block from negibouring correct macroblock for the block with highest priority. If the matching failed, we will use structure reconstruction method to repair the 4x4 block, using the information of the neighbouring block to inpaint each point of the 4×4 block by summing the estimates of neighbouring points, weighted by a normalized weighting function. For temporal error concealment, we modify the motion inpainting algorithm, using the motion vector of the correct received macroblock with different order according to the conditions of neighbour macroblock. From the experimental results, the algorithm proposed in this thesis get better results than the original error concealment algorithm, both in the subjective evaluation criteria and objective evaluation criteria.