No-reference Natural Image Quality Assessment

About three-quarter of amount of information obtained by human is extracted from images in the outside world. With the development of computer science technology, image engineering has been widely applied to every domain of national economy. However, although there are great improvements in image technology domain, the compromise between image processing algorithm and facility design is still needed in practical applications, such as that between spatial resolution and image dimensions. The visual quality of reconstructed image then would be affected by the compromise. Therefore, in order to get the ultimate results, it is necessary to know what and how the influence is. Employing objective metrics of image quality assessment, we could evaluate the image processing method effectively and ultimately get better visual quality. Among all kinds of quality metrics, no-reference image quality assessment methods had gained extensive and in depth research, because it does not need any reference informantion and could be directly applied to the terminal of systems.This paper aims to deeply and systematically study the No-Reference Natural IQA (NR-NIQA). At first, a brief introduction of the wavelet and statistical characteristics of natural image in wavelet domain is presented, and then a wavelet domain statistics based NR-NIQA method is proposed. Secondly, wavelet is expanded to contourlet which could represent image more effectively. Based on the natural image statistics in contourlet domain, the image quality features could be modeled precisely, and then a contourlet domain statistics based NR-NIQA method is proposed. This algorithm describes the quality features of natural image from the point of multiresolution and multidirection, and represents visual quality by capturing the influence of distortion on the relationship of contourlet coefficients. Experimental results show that it has a good consistency with subjective evaluation. Finally, image modeling in transform domain is expanded to hidden Markov model, and a wavelet domain HMT based NR-NIQA method is proposed. It employs the powerful mathematical tool——HMT to capture the relationship of wavelet coefficients and represent image visual quality. Experimental results demonstrate the feasibility and validity of it.