Digital Image Quality Assessment Algorithms Research And Performance Assessment

With the development of multimedia technology and network technology, many kinds of terminals like camera, mobile phone, projector, and TV has spread rapidly among the people, which makes the amount of images in our life increased immensely. In the process of transmission, compression distortions and transmission distortions inevitably cause the decrease of the image quality, and will trouble the terminal users a lot because of serious distortion. Therefore, the assessment of image quality is very important. Though the image quality can be derived correctly from subjective assessment, the long time consuming, high cost and human involvement make this method inconvenient to copy and promote. PSNR is widely used because of simple calculation, but its accuracy is still disappointing. In the past three decade, research about image quality assessment have been trying to propose an algorithm that displays better performance than PSNR, and can replace PSNR in technical application, so a lot of algorithms have came out. The goal of this paper is to propose a novel IQA algorithm that will perform better.It is proposed in this paper that the performance evaluation of several IQAs is an important basis of adjusting them. Therefore, one part of this paper’s work is to put forward a set of project to assess the performance of IQA. In this project, the followings are included: TID2008 image database, curve-fitting of the algorithm’s metric data and the calculating of performance measurement. Experiments have been conducted under PSNR, SSIM and MSSIM algorithms. The experiment results demonstrate that there is still big difference between the existing IQAs and human subjective assessment, and their performances still have potential to improve.Besides performance assessment experiments, this paper also proposes an improved algorithm TSSIM on the basis of SSIM. This new algorithm uses the feature of Contourlet transform that can resolve an image in different frequencies and different directions. The final and single metric is achieved by more analysis of SSIM map. There are six design details proposed in this paper. As a result, the one with the best performance become the final TSSIM. The evaluation index of the final TSSIM shows that, when it comes to qualify the image fidelity, it is not only reasonable to set different weight to different frequency subbands, but also accords with the rule of human perceptual to divide the information of an image into luminance, contrast and structural parts, and then combine them together in some certain way.