A Video Quality Assessment Based On Saliency

With the rapid development of computer and network communication technology, the need of people for multimedia information such as pictures, videos becomes stronger. Video compression and transmission process tends to cause a decline in the quality of the video, and the video quality can reflect the quality or performance of the system or the transmission channel. Therefore, the need for video quality assessment is necessary. Person is the ultimate recipient of video applications. Therefore, people have conducted subjective evaluation based on the evaluation rule is the most accurate way. A number of international standards for subjective testing have been established to guide people to the subjective test. However, subjective evaluation needs to spend a lot of time and effort. Also, it can’t be embedded into the video application system. In recent years, people have studied a number of objective video quality evaluation methods. Objective video quality assessment can be divided into three categories: full-reference evaluation methods (Full-Reference, FR), part of the reference evaluation methods (Reduce-Reference, RR), no reference to the evaluation method (No-Reference, NR). Full and partial reference evaluation methods are dependent on the original video, and in many cases, the video receiving side is difficult to obtain the original video. In contrast, no-reference methods do not require any information about the original video. They have better flexibility because they only need existing video to evaluate video quality. Therefore, the study of no-reference video quality assessments have became a hot spot. Video Quality Experts Group (VQEG) believes no-reference video quality assessment will be one of the focus of future research directions. At the same time, more and more people began to study the existing no-reference video evaluation algorithm, which were based on the introduction of the human visual characteristics, the establishment of appropriate visual model to improve the accuracy of the evaluation.This article has researched many no-reference video evaluation methods, and proposes a no-reference evaluation method based on Peak Signal Noise Ratio (PSNR) and blockiness. The method is based on PSNR which have used in full reference evaluation. Firstly, it need to detect the intra-coded frames which are used as the original frame of the distorted video and use its adjacent frames as distorted frames, then calculating their PSNR; secondly, blockiness value is calculated from the distorted video and it is weighted with the peak signal to noise ratio; finally, to determine their value factor by fitting the right. This paper compares the algorithm with some existing video evaluation algorithms, experimental results show that the algorithm results are closer to the subjective evaluation results, the correlation is better.This paper also studied the human visual features and some significant model. At the same time, the visual model incorporated into the calculation of PSNR and blockiness and the no-reference evaluation method based on saliency is proposed. Firstly, in the conventional saliency model, this paper chose two models:model based on the spectrum analysis and cognitive model as the way to get the saliency map of video’s every frame, the significant value of each pixel as weights; Secondly, the algorithm also considers the human masking characteristics to improve the model of weights, so the weight is more in line with the subjective characteristics of the human eye; finally, improve the calculation of PSNR and blockiness. This paper compares the algorithm with some existing video evaluation algorithms, experimental results show that the effect of the algorithm is better compared with the original algorithm.