Research On Image And Video Quality Assessment

With the popularity of computers and mobile Internet, digital images and video applications have been developed rapidly. Digital images and video, which contains a lot of useful information, is an important medium for human and have been widely used in medical, remote sensing and military fields. However, in the process of capture, compression, storage, transmission and display, there always produce noise inevitably. As a result, great obstacle is brought to the image and video understanding. Measuring the degree of distortion of images and videos is very important for compression algorithms and optimizing the performance of network transmission. Traditionally, image and video quality evaluation always uses subjective quality assessment strategy for its accurate prediction. However, these method are turned out to be time-consuming and easily affected by environment. Therefore, objective quality assessment method which can automatically assess image and video quality is desired. According to the dependence of original image and video, objective image and video quality assessment methods can be divided into three categories:full-reference, reduced reference and no reference. In full reference quality assessment, a complete reference image or video is to be known. However, a lot of images and videos applications do not have any reference image or video. So there is a need for no reference image and video quality assessment, but no reference evaluation has a lower prediction accuracy compared with full reference methods. No reference and full advantage of the comprehensive reference, semi-reference quality assessment in the original image to extract some useful information as a secondary evaluation. Reduced reference image and video quality assessment are widely used where only partial information of reference image and video is available. In this paper, we focus on full reference and the proposed metrics are summarized as follows:(1) Motivated by observations that human visual system is highly adapted to extract irregularity information of textures in a scene, we introduce multifractal formalism into image quality assessment scheme. Based on multifractal analysis, statistical complexity features of nature image are extracted robustly. Then a novel framework for full reference image quality assessment is further proposed by quantifying the discrepancies between multifractal spectrums of images. Experimental results demonstrate that the proposed method is highly consistent with human perception.(2) Motivated by human visual perception is sensitive to orientations and detail texture information, test video are decomposed by Gabor filter to capture the salient visual properties. Then the local binary pattern operates on each sub-band to extract local neighbor structures information. Finally, video quality is quantified by the statistical features from all the sub-bands combined with a motion perception model. Experimental results show the proposed method is outstanding when compared with state-of-the-art metrics.