A No-Rrference Image Quality Assessment Algorithm Based On Region Of Interest

Since Daguerreo invented the camera in 1839,images have played an extremely important role in human's life.With the development of the Internet and electronic equipments,the ways of images' making,transmissing and displaying are becoming more diversified.At the same time,degradations are introduced,which may occur due to presence of noise,blocking artifacts,blurring,fading etc,and make images lose their detials or informations.Therefore,image quality assessment(IQA)has become more and more important nowadays.Image quality assessment is able to automatically predict the quality of digital pictures by building the similar model of human visual system(HVS).In this work,we introduce a spatial-domain no-reference image quality assessment(NR-IQA)model based on cutting the image's region of interest.Firstly,we analyze a classic visual saliency model and make a improvement of it,use it to divide the image into the region of interest and other area.Then we use existing no-reference image quality assessment technique to extract the feature of distortion.Finally we exploit a support vector machine(SVM)to learn the mapping function from those features to the perceived quelity scores and build an NR-IQA model.We make experiments with different database and different types of distortion,the result shows the new model predicts the quality of images more accurately than most of other NRIQA models and has a high agreement with respect to human subjective scores.