Research On Correlation Noise Modeling For Distributed Video Coding

Distributed video coding (DVC) schemes encode video frames independently at the encoder and decode them jointly by taking advantage of the correlation between frames at the decoder. Compared to the traditional video coding standards, such as MPEG-X and H.26X, DVC transferred the computational complexity of the encoder to the decoder side. Due to the characteristics of simple encoding, DVC is sutibale for mobile video phone, wireless video surveillance and wireless video sensor networks, in which computing power and power consumption of wireless terminal is constrained.This thesis focuses on correlation noise modeling (CNM) for DVC. And some meaningful research results have been obtained as following.Firstly, a novel fitting method of distribution parameters for correlation noise models based on minimum Euclidean distance is presented. The fitting approach can obtain the differences between the Laplace distribution and the correlation noise with Euclidean distance. And then it can determine accurate Laplace coefficients by calculating the minimum Euclidean distance. This thesis analyzes four fitting methods for probability distribution of the correlation noise and compares the performance of them through experiments. These fitting methods includes:minimum Euclidean distance, variance, central moments and maximum likelihood. The average value of peak signal noise ratio (PSNR) by applying the proposed method is higher than other fitting methods about1.73dB under the same testing conditions. And the presented fitting method can improve the rate-distortion performance of DVC successfully.Secondly, a novel CNM algorithm based on multiple probability distributions for DVC has been proposed. The proposed CNM method can analyze the probability distribution features of the DCT sub-bands adaptively and select the best suitable probability distribution according to the similarity of entropy between the sub-bands and the three candidate probability distributions, including Cauchy, Laplace and Gaussian. Experimental results show that compared with the existing typical CNM approaches, the proposed CNM method can improve the rate-distortion performance of the DVC system significantly for whatever offline or online manner.Thirdly, a scheme of correlation noise estimation based on wireless transmission applications for DVC has been presented. This scheme is able to conceal the transmission errors which occur in the macroblocks of keyframes and use two keyframes to estimate the correlation noise at the decoder. In order to reduce the impact of errors by the keyframes and the errors of correlation noise estimation, the proposed schem use the approach based on experience threshold to correct the correlation noise. Experimental results show that the proposed approach can ensure the performance of the DVC effectively.