Visual Perception-based Image And Video Quality Assessment

Objective quality assessment (QA) plays an important role in relative application fields of image and video. During the period of capture, compression, storage, communication and transmission, image/video will inevitably suffer distortion more or less. It is of great necessity to complete an algorithm which can evaluate the distortion or fidelity of the image/video objectively, efficiently and simply. During the period of application, a good QA metric can be employed to guide the construction and adjustment of processing systems, and to optimize the processing algorithms, or to benchmark image and video processing systems and algorithms.As the terminal for receiving the visual information, the human eye is the most precise instrument to assess the quality of images or videos. Then, it could improve the performance of algorithms if we take the characters of the human visual perception into account. For the image quality assessment with FR (Full Reference) model, the thesis proposes a perception and content based quality metric, which is on the basis of SSIM (Structural Similarity) and makes full use of three essential factors of distortion, the"amount of error", the"location of error", and the"structure of error". Most importantly, this metric utilizes the human subjective perception to accomplish the quality evaluation. The results of the experiments demonstrate that the correlation between the objective scores and the MOS (Mean Opinion Scores) is improved obviously. At the same time, this algorithm is used to assess the video quality, and the obtained performance is better than that of the ten models validated by VQEG (Video Quality Expert Group).Aimed to the NR (No-Reference) QA of images compressed by JPEG, the thesis proposes two metrics within the transformed domain. The first one is finished in DCT domain, which only partly decodes the stream and then calculates the percentage of zero coefficients. And it takes the CSF (Contrast Sensitivity Function) to weight all the percentages at different frequencies. Because this method is not necessary to completely decode the image, it can dramatically increase the speed of computation. The second evaluates the image quality in Wavelet domain, which completes the QA procedure by inspecting the change of statistical features of Wavelet coefficients. The experimental results show that the performance of these two algorithms is better than that of the FR model of PSNR.At last, the thesis extends the video database released by VQEG. In order to evaluate the distortion produced by video compression with standard MPEG-4 and H.264, the thesis constructs a new video database. All of the videos in this database are compressed by MPEG-4 and H.264 standards and assessed by human viewer. This video database offers a good beginning for the future video quality assessment research.