Blind Image Quality Assessment Based on Natural Scene Statistics

Blind measurement of visual quality is of fundamental importance in numerous image and video processing applications. Most of the no-reference Image Quality Assessment (NR IQA) methods are distortion-specific and their application domain is limited. Also, almost all distortion-generic NR IQA are computationally complex, making their applicability in real time applications very limited. In this paper fast blind distortion-generic IQA is proposed. This method uses natural scene statistics of normalized luminance coefficients. This would quantify possible losses of ‘naturalness’ that are caused by the presence of distortions. The best relevant sources of distortion are selected and fed to an Artificial Neural Network. The blind method is tested on the “LIVE” dataset. Experimental results show that our blind method correlates highly with subjective quality assessment results. Also this blind method has a very low computational complexity that makes it very appealing for real time applications.