Research On Video Quality Assessment Algorithm Based On Wavelet Transform

With the rapid development of digital networks, the digital video technology makes big difference to real-life, such as video surveillance, video conferencing and digital cinema. In the process of the video capture, enhancement, compression, transmission, transcoding, reconstruction, recovery, etc. Due to the video capture devices, mass storage devices, processing algorithms, and transmission equipment, it cannot be avoided to bring distortion effects such as blur, blocking artifacts and noise thereby reducing the quality of the video. Such as blocking or blurring distortion in the process of video compressed and coding and error generated in the network such as the transmission delay, jitter, packet loss and other distortions, it will result in lower video quality. It is necessary to judge the level of the video quality before it is applied in order to save time and cost. Because most direct receptor to video and image is the human eye, so now people are more concerned about how we can design video quality assessment evaluation model that make the subjective judgment is in consistent with objective judgment, which highlight the prominent importance of the human visual characteristics of the visual system. The local, multi-resolution and directional characteristics of the wavelet transform is similar to the characteristics of the human eye to a certain extent, so based on this similarity, the paper will in the wavelet domain to study video objective quality assessment algorithm. The main research direction in this article is for the full reference video quality assessment, content and results of the study are as follows:1. Propose a video quality assessment algorithm based on wavelet domain and time-domain.The video image from the zoning method, wavelet transform, Euclidean distance, Chebyshev distance, Manhattan distance and time domain fused these directions to construct video quality assessment algorithm. First, the sensitivity of the human eye to make an image of the different regions are different, the video image area divided to two parts, and select different weighting weights to the different regions; Next, extract high-frequency coefficients and low frequency coefficients generated by wavelet transform to constitute a feature vector; Then use the Euclidean distance, Chebyshev distance and Manhattan distance to calculate the feature vector distance, to represent the similarity between the reference video and distortion video; Finally, use the human eye perceptual model for successive video frames for time domain to process of single frame quality, the obtain the video quality score. Experimental results on LIVE database show that the algorithm when using Euclidean distance, distance superior performance than the other two.2. Proposed a video quality assessment method based on low-frequency coefficients of wavelet combined with significant information. The algorithm combines a wavelet transform, motion estimation, and other factors significant information to design video quality evaluation algorithm. First, use the low-frequency coefficients generated from reference video and video distortion by wavelet transform to measure low-frequency distortion of the video. Next, compare the reconstructed image from the high frequency coefficients and the saliency map extracted from video image,the results show that video image saliency map have the better performance in the evaluating the high frequency distortion.Then, use the motion vector extracted from a reference video to do the motion estimation and motion compensation to the reference video and distortion video to obtain the weighting factor of each frame, and then weight the quality of the video frame; Finally, time domain processing of successive video frames is used to give the video quality value. From the experimental results, the performance of this algorithm is more superior, and has the high consistency with human visual system.