Image Quality Assessment By Relative Gradient Statistics And Contrast Masking

Given the exploding development and deployment of visual acquisition, display and processing technologies, digital images have become ubiquitous in our daily life. The storage, transmission or other processes of digital images often bring the image distortion into the image which can cause the decline of image quality. In that case, the problem that how to analyze and predict the image quality is a hot topic in the image processing field.Image quality assessment mothods can be divided into subjective methods and objective methods. However, the subjective methods cost too much, and can be affected by personal or other factors. Thus, the objective evaluation methods get more attention from researchers. In this paper, we utilize relative image gradient and contrast masking to study the existing objective methods. The main work is shown as follow:(1) Blind image quality assessment by relative gradient statistics. Because that the traditional image gradient information can not describe the degree of image distortion accurately, this paper proposes a no-reference image quality assessment method which utilizes the image gradient information and relative gradient information. In this method, we utilize the gradient relative variation degree to represent the traditional image gradient information, and use the correlation information between the adjacent gradient to enhance the accuracy of describing image quality. Meanwhile, the gradient magnitude, the relative gradient orientation and the relative gradient magnitude are used together to represent the image gradient information, and the variance of their histograms is considered as the image quality features. At last, we use BP neural network with AdaBoost algorithm to build the mapping from image features to image quality. Experimental results show that this algorithm has robustness performance.(2) Medical image quality assessment via contrast masking. In order to expand the application of image quality assessment in medical image, this paper proposes a medical image quality assessment method based on the contrast masking theory. First, according to the contrast masking theory, the image can be divided into four regions, which have the different visual sensitivity weights. Then we form the structure similarity metric to quantify the image distortion degree in four regions, utilizing the image foreground brightness, image background brightness and the just noticeable difference(JND). We testify this method in LIVE image quality assessment database and our designed medical image quality assessment database. Experimental results show that this method has high subjective consistency.