Video And Image Quality Assessment Based On Human Visual Perception

Recent years, with the rapid development of broadband digital networks and theadvent of VLSI technology as well as the high-speed multimedia special chip, all kindsof multimedia technologies have been widely used in real life. The digital image andvideo have been widely applied to all aspects of life, such as video surveillance, videoconference technologies and video on demand as their vivacity, convenience andrichness of content. However, during the acquisition, storage, processing andtransmission process, the video quality will be inevitably decreased due to theacquisition system, storage medium, processing algorithms and transmission equipment.The degradation of the image and video quality can reflect the performance of thesystem or the Quality of Service (QoS) of the transmission channel. Thus, it is of greatpractical significance and application value to evaluate the image and video quality.From the characteristics of human visual perception to video quality, according tothe effects of various distortion types on the video and image quality, and aiming toexplore the relation between human visual system and video and image quality, theobjective video and image quality evaluation models are proposed, summarized asfollows: 1) According to the origin of distortion of compressed images, an image qualityassessment metric is proposed, which is based on Directional Discrete CosineTransform (DDCT). By measuring the loss of detail information as well as change ofdirections, and considering the human eye's perceptual characteristics to luminance, aNo Reference quality assessment metric for compressed images is designed. Theexperimental results show that this metric has high consistency with subjectiveperception. 2) Based on feature extraction and sparse representation theory, andaccording to the visual perceptual characteristics, a 3-D wavelet based video qualityassessment metric is proposed. Firstly, 3-D wavelet is employed to obtain the videofeatures for simulating the multi-channel characteristic of human visual system. Then,the coefficients are filtered by contrast sensitivity function according to human visualpsychophysical and masking characteristics. Finally, to model the human eyes'perceptual process more accurately, the short-term memory characteristics areconsidered to incorporate the quality of each frame, and then the overall video quality isobtained. Experimental results show that the proposed metric can reflect the subjectiveperception to video quality accurately. 3) According to the characteristics of the physiology and psychology of human visual system, a noval visual attention modelwhich can be used in video quality assessment is proposed. Video content is extractedthrough computing the gradient of video frames. Visual attention on motion objects ismodeled by employing the relative motion vector. Visual perceptual uncertainty iscomputed by exploiting the absolute motion vector. Finally, visual attention saliencymap is obtained by incorporating the three factors described above. Experimental resultsshow that this model can improve the existing classis metrics when employing it asspatial-temporal weighting factors.