Research On Error Concealment In Video Communication Based On H.264

With the fast development of electronic technology, computer science, telecommunication and theory of video compression, video communication is one of the main applications in multimedia communication. At present, the most wired and wireless networks can’t provide QoS guarantee, so the error and loss of video data during transmission are inevitable. The latest video coding standard H.264 is provided formally by ITU and ISO. With higher coding efficiency and better network friendship, H.264 has become the primary video coding standard applied in video communication. But the video stream compressed by H.264 is more sensitive to transmission errors. Once bit error or packet loss occurs in transmission, the effects of fault information may be propagated in spatial and temporal rapidly which can significantly degrade the image quality. To enhance video performance in error prone environment, many techniques have been proposed, which can be divided into three categories:error resilience encoding, error concealment and interactive error compensation. This paper studies error concealment technology in video communication based on H.264, which efficiently recovers the corrupted or lost video data under the condition that the recovered image must preserve the original contents and satisfy the human visual characteristics. The main work and contributions in this thesis are as follows:1. A novel method estimating the neighboring pixels which is the strongest relevant to border pixel is proposed, which exploits the same texture features within the local region. For a lost border pixel, a region with 5×3 adjacent pixels forms a check window. According to the position relationship between pixels, the check window is divided into seven directions. The difference between pixels belong to the different directions is counted respectively. The direction with the minimal difference is edge direction of check window. The neighboring pixel which has the edge direction with the lost pixel is the strongest relevant pixel. Then, a new spatial error concealment algorithm using minimum variance of edge of border pixel is put forward. On the basis of fundamental pixel interpolation, the model parametersωΥ,ωB,ω1 andωR are introduced to constitute the equations to recover the lost pixels. Based on the principle of minimum variance of edge of border pixels, the model parameters are obtained. At last, the all lost pixels are recovered by the equations. The simulation results show that the proposed algorithm can not only guarantee the boundary smooth and eliminate the block artifact, but also preserve the partial edge information of the lost block.2. To overcome the inaccurate estimation of motion vectors in temporal error concealment, a motion vector recovery method based on the minimum variance of boundary of motion vectors is presented. On the basis of distance-weighted algorithm, the parametersωT andωB are introduced to constitute the expressions to estimate the lost motion vector. Based on the principle of the minimum variance of boundary of motion vector, the parameters are solved. The lost motion vector is recovered through the expressions. Furthermore, a novel temporal error concealment algorithm based on relative strength of motion vectors (RSMV) is proposed, which efficiently exploits the spatial correlation of motion. For a lost block, the RSMV of its neighboring area is first calculated according to the motion vectors of adjacent blocks. Then, either the minimum variance method of boundary of motion vectors or Lagrange interpolation method is employed to recover the lost motion vector according to the magnitude of RSMV. The simulation results show that the proposed algorithm constantly outperforms the traditional algorithms and demonstrates fairly good robustness.3. A motion vector recovery method based on the characteristics of similar triangle is given, which exploits the fact that the corresponding blocks of successive frames have similar motion tendency. Firstly, the horizontal component, vertical component of motion vector, and the frame timestamp can be used to constitute a 3-dimension space. Then, the lost block, the spatial neighboring block and the temporal neighboring block are projected into the space. The motion vectors can be connected as similar triangle. According to the character of similar triangle, the lost motion vector can be recovered. At the same time, a novel temporal error concealment algorithm is proposed, which exploited the temporal and spatial correlation of motion. For a lost macro block, the temporal related motion vector and spatial related motion vector are first estimated by similar triangle method and Lagrange interpolation method respectively. Then the temporal and spatial correlation of motion is calculated. The recovery motion vector is obtained by using a weighted average of the temporal and spatial related motion vector according to the magnitude of correlation. Finally, variable block size is adaptively selected according to the similarity of motion vectors of adjacent blocks and the features of H.264 to conceal the lost macro block. The simulation results show that proposed algorithm can recover images with high quality in different rates of lost block or different video sequences.4. To overcome the block artifact or the rough motion caused by concealment algorithms using uniform block size, a new frame error concealment algorithm using adaptive block sizes is proposed. Firstly,4x4 blocks of the adjacent frames are extrapolated onto the lost frame and the overlapped areas between motion extrapolation areas and lost frame are calculated. The extrapolation direction is selected according to the size of overlapped areas and the frame type of adjacent frames. For a lost macro block, the neighboring macro blocks of adjacent frame are extrapolated along the extrapolation direction, either macro block size or 8x8 block size is adaptively selected to conceal the lost macro block based on the size of overlapped areas and the similarity of extrapolation motion vectors. For an 8x8 lost block, either the 8x8 block size or 4x4 block size is selected in the same way. Other blocks are reconstructed by outer boundary matching algorithm. The simulation results exhibit that the proposed algorithm can outperform the conventional algorithms in terms of both objective and subjective video quality. On the other hand, the computational complexity of the proposed is moderate, which leads to its wide practicability.