| During recent years, great progresses have been made due to the effort of world-wide scholars. Nevertheless, with the development of other areas in image processing, image quality assessment models could perform even better.The thesis will pay attention to full-reference image quality assessment, and evaluate the several increasingly popular models. Then, two novel full-reference image quality assessment models will be proposed based on SSIM. The detailed information is as follows:1. By evaluating the four dominating quality models, this thesis will describe the avantages and disadvantages of these existing IQA models, and will provide detailed informations about LIVE database and natural scene statistics. By analizing the features of Pearson Correlation Coefficient (CC) and Spearman Rank Order Correlation Coefficient (SROCC), this thesis will point out that CC is more important than SROCC on evaluating IQA models.2. By analyzing the feature of human eyes based on daily experiences, this thesis will point out that human eyes tend to focus on what they are interested in and neglect the other parts. According to the conclusion, a strategy will be proposed and SSIM will be modified based on it (ROI-SSIM). Then, a series of experiments on LIVE database and natural scene statictics will be used to prove that the performance of ROI-SSIM will be improved.3. By thorough analysis on Contourlet transformation, including its features and every process, this thesis will seek a balance point between complexity and accuracy of algorithms and select proper laplacian pyramid layers and directional filter bank stages. Give every layer and its stage a corresponding weigh according to human eyes are more sensitive to parts with lower frequency than higher ones. Then, by combining Contourlet and SSIM, a novel IQA model (CSSIM) will be proposed. Experiments on LIVE database and natural scene statictics will be used to prove that the performance of CSSIM will be improved. |
PDF Full Text Request