Research On Image Quality Assessment Based On Visual Regions Of Interest

Researches on image quality assessment are meaningful and basic for the imageprocessing engineering. Traditional image quality assessment based on error sensitivity isclear on physical meaning and simple on calculation, but the results of them have a littledifference with feeling of human vision. In recent years objective evaluation which was putforward based on error algorithm of image structure consistent with visual subjective feelingsin a certain extent, but only image structure distortion was considered without considering thehuman visual characteristics, which ignore the important role of people in image qualityassessment.Although the perspective is very wide when people observe images, their concentrationrange is very small. They pay more attention on attention points. Compared with other regions,the distortion in this part has a very large impact on image quality. Based on the deep researchon image quality assessment method and the structural similarity theory, this paper proposes amethod of image quality assessment algorithm based on visual regions of interest according tothe characteristics of human vision. Combining attention mechanism of human visual system,the algorithm which simulates the understanding of image region of interest obtains a methodof algorithm. Combining with the existing structure similarity algorithm, this paper gets thedegree of interest quality evaluation method algorithm based on the degree of interest factor isfused degree of interest of image quality in different regions. In order to obtain better effect ofsimulating multi-channel feature and visual sensitivity feature of human vision. The referenceimages and those to be tested are decomposed into different sub-band. Finally, image qualityevaluation results, combining with the structural similarity and degree of interest factor basedon the degree of interest in different sub-band image assessment and contrast sensitivityfunction pairs band image quality weighted, is obtained. A large number of experiments on the LIVE image database verify the correctness andthe sensitivity of the algorithm proposed in this paper. The test results keep a goodconsistency with subjective feelings.