LAUSR

Image is a typical example of visual data, and its quality inevitably affects its application. Hence, measuring image quality accurately is a beneficial task. In practical application, there are different image types e.g. natural image and screen content image (SCIs). And the distortion types contained in images are various. Most image quality assessment (IQA) methods concentrate on a single image type with limited distortion types. In this paper, we present a no-reference IQA method which can accurately measure the quality for both natural image and SCI, and is robust for various distortion types. Human visual system is sensitive to the changes in image structural information which are usually caused by image quality degradation. Therefore, the new method employs local structural information representation for IQA. We first analyze the gray-scale fluctuation of each pixel in four detection directions to obtain four gray-scale fluctuation maps (GFMs) and one gray-scale fluctuation direction map (GFD). And then, the structural features extracted from GFMs and GFD are used for representing local structural information. Finally, the mapping function from the features to image subjective scores is trained by support vector regression (SVR). The experimental results on the public databases demonstrate that SVR is suitable for IQA and the proposed method can accurately predict the quality of both natural images and SCIs with various distortion types.