No-Reference Image Quality Assessment Based On HVS

The process of the collection, process, transmission and storage of image often leads todegradation or distortion, this problem obstructs the development of perceiving object thingsand research of image information. Hence, reasonable assessment of image quality isimportant. However, some distortions haven’t the original image to refer. So no-referenceimage quality evaluation emerges. According to current research status, no-reference imagequality evaluation methods include algorithm based on distortion and algorithm based onmachine learning. This paper studies no-reference method by these two algorithms based onJPEG.In this paper, the algorithm based on distortion aims at blocking effect. This paperanalyses visibility coefficient due to texture masking, realizes that when caclulating thebackground activity, the method Laws proposed is good, but this wastes a lot of time.According to this problem, this paper proposes modified visibility coefficient due to texturemasking, Visibility Coefficient due to Blocking Metric(VCBM). Experimental indicates thatmodified method not only reduces a lot of time, but also has a good fitness with DegredationMean Opinion Score(DMOS) than original method. Further more, compared with otherfamous methods, modified method achieves a better result.This paper also proposes a JPEG no-reference image assessment based on ε－SVR,Support Vector Regression Objective Score(SVROS). It combines Human Visual System(HVS) with blocking metric effectively, extracts luminance contrast, texture complexity,block artifacts and block flatness. In consideration of luminance can refect block flatness andtexture can influence block artifacts, it proposes luminance contrast weights block flatnessand texture complexity weights block artifacts,then an integrated feature is obtained. The finalobjective quality assessment result, SVROS, is predicted by ε－SVR. The experimental results indicate that the extracted feature fits HVS well, and the objective scroe agrees wellwith DMOS.