Research On Source Distortion Estimation For Distributed Video Coding

Distributed Video Coding(DVC)is a video coding technology based on the theories of Slepian-Wolf and Wyner-Ziv(WZ).Compared with the traditional video coding,DVC shifts the complex motion estimation and compensation form the encoder to the decoder.Hence,DVC is suitable for the emerging video applications,such as wireless video surveillance and wireless video sensor networks and so on.In a wireless video communication system,accurate distortion estimation of the reconstructed videos is significant to improve the system performance and optimize the resource allocation of the system.Although there have been some methods for the distortion estimation of DVC,they cannot be suitable for real-time applications since the original videos information need to be provided.This thesis focuses on no-reference source distortion estimation for DVC.And the results obtained are as follows.Firstly,a no-reference source distortion estimation algorithm based on Stanford reconstruction for DVC is presented,which only utilizes the information at the decoder instead of the encoder side.Thus,the proposed method can perform the video distortion estimation without increasing the computational complexity of the encoder.In the presented method,the distortion incurred by the quantization and reconstruction is taken into account.Besides,the estimation method of Laplacian distribution parameters is also developed to improve the accuracy of distortion estimation.To evaluate the performance of the proposed algorithm,the group of picture(GOP)are set as 2 and 8,respectively.And the experimental results show that the results of the proposed algorithm is more closer to the practical distortion values than that of the existing algorithm.Specially,the absolute errors between the practical values and the estimated results are less than 1.5 dB,the relative errors are within 5%.As the GOP are set as 2,the minimum absolute errors is 0.03 dB,the minimum relative errors is 0.09%;while GOP are set as 8,the minimum absolute errors is 0.02 dB and the minimum relative errors is 0.07%for different video sequences.This indicates that the proposed method can achieve a good result for video distortion of DVC.Secondly,the performance of the minimum mean squared error(MMSE)reconstruction algorithm is superior to other reconstruction algorithms in the existing DVC scheme.However,there are no source distortion estimation models based on MMSE reconstruction have been proposed.In this thesis,a no-reference source distortion estimation algorithm based on MMSE reconstruction for DVC is presented,in which the correlation between the original WZ frames and the side information have been explored.Experimental results show that the absolute errors between the practical values and the estimated results are less than 1.5 dB,and the relative errors are within 5%for different video sequences.As the GOP are set as 2,the minimum absolute errors is 0.01 dB and the minimum relative errors is 0.03%;while GOP are set as 8,the minimum absolute errors is 0.01 dB and the minimum relative errors is 0.03%.This results show that the estimation accuracy of the proposed no-reference source distortion estimation algorithm based on MMSE reconstruction for DVC is very high.