Image Quality Assessment Algorithms Research Based On MMTD

Various kinds of intelligent terminal and display appear more and more frequently in people’sdaily life, the image can be found everywhere as the most common and effective information carrier.If the image quality can’t be guaranteed, will not only affect the user’s visual experience, and evenaffect the performance of the practical application. Therefore, assessment of image quality is veryimportant. There are many disadvantages in subjective assessment, such as long period, high costand tougher environmental requirements. It is not suitable for system integration and industrialpromotion. Objective quality assessment is more suitable for industrial application. A lot of effortsis required to research on image quality assessment and target at present a method with performanceclose to the subjective assessment This paper proposes two objective image quality assessmentmethods. The main work:(1)The assessment system for performance of image quality assessment methods is studied inthis paper, the analysis and implementation of the quantified index is emphasized, the effect ofthree classical image quality assessment methods have been assessed based on the system. Theresult verifies that the assessment system is effective and accurate, so it can be used as the standardto assess the improved IQA method.(2) MMTD-SSIM is proposed in this paper as an improved method of structure similarity. Themethod introduces visual attention model with which human identification the image into the regionof interest, and uses theory of MMTD to determine the local salient degree as the weight of eachregion. The final score of the new method is obtained by integrating all the regions of weight.Experiment results show that giving higher weights to regions of interest approach better resultswhich is more consistent with subjective assessment results.(3) CTSSIM is proposed in this paper.It is based on structure similarity method andContourlet transform. Contourlet transform reflects the characteristics of the image feature better,and it can properly describe the smooth contour. Its advantages indicate that it is a more advancedimage representation method. The best decomposition scheme is determined by experiments. Theimproved method uses Contourlet transform tool for multi-scale decomposition of image, and usethe MMTD combined with SSIM map image information measure the distortion degree which value is the weight of the subbands. The different scales of frequency information have been treateddifferently. Experiment results show that the improved method has good consistency with thesubjective assessment results, which illustrates the superiority of the multi-scale analysis.