Research On Video Quality Assessment Based On Feature

With the continuous development of multimedia technology and the increasing number of network users,digital video has become an important tool for people to get information,such as video learning,Internet TV and so on.However,processes of video information,such as video compression,transmission and reconstruction will cause a decline in video's quality.Therefore,the use of an accurate and effective video quality assessment method has attracted more attention.The evaluation method is divided into subjective and objective assessment.Subjective assessment has been recognized as the most reliable method.However,subjective assessment wastes a lot of time and needs a large amount of manpower.So the objective evaluation method is particularly important.Here are some research results about the objective quality assessment.The main contents are as follows:(1)A no-reference video quality assessment model that utilizes many spatial and frequency features.This method extracts many perceptual features,including gray-level gradient co-occurrence matrix,spatial entropy,spectral entropy,correntropy and a natural index features.Finally,we train these features by support vector regression to build the relationship with perceptual features and quality of distorted video.The proposed model is tested on LIVE and IVP VQA database and results prove that our algorithm can achieve much better performance than state-of –the-art published algorithm.(2)A no-reference video quality assessment that utilizes visual salient features.Firstly,extract the saliency map,and the significant value of each pixel is taken as the weight to multiply the video frame difference to obtain the video frame difference with the significant region,and then extract distorted features on these video frame differences.Secondly,VS is used as a feature when predicting the quality of distorted videos.Finally,we train these features by SVR to build the relationship with features and quality of distorted video.The experiment results on LIVE and IVP VQA database prove that our algorithm can achieve higher consistency with the subjective evaluation.(3)A no-reference video quality assessment combining DCT transform and motion information.Firstly,we divide video frame and video frame difference into groups by analyzing the time domain characteristics of the video,extracting shape features from DCT coefficient of each group,and then we get the features of a whole video with temporal pooling.Secondly,we train these features by SVR to build the relationship with perceptual features and quality of distorted video.The experiment results on LIVE and IVP VQA database prove that our algorithm can achieve higher consistency with the subjective evaluation.