A New Comparison Method for Full Reference Image Quality Metric

Image quality assessment becomes more and more important with the development of image and video processing applications. Over the years, many new image quality metrics have emerged. It is important to evaluate performance of these quality metrics under the same conditions and analyze their strengths and weaknesses. In this paper, we first review others´ work, analyze the characteristics a good metrics should satisfy and propose a philosophy based upon our analysis. Then we design a new system which reflects metric´s characteristics by comparing curvature array of metric function and execute it to degraded image. 3 kinds of point-spread functions: Gaussian, motion and disk, and 3 kinds of noise: Gaussian noise, salt noise and uniform noise were considered in our experiment. Our experiment results have shows characteristics of metrics and exhibited some interesting conclusions. These conclusions might be helpful in designing a new image quality metric.