The Research On Image Blur Assessment

With the rapid development of information technology, multimedia technology has been widely applied to every corner in our everyday life. However, current multimedia systems are not perfect, and the limited network capacity can not fulfill the needs of multimedia transmissions. Therefore distortions are inevitably introduced for actual image and video through capture, compression, transmission and storage. Among all kinds of distortion, blur is one of the major representations of image and video quality degradation. Assessment of image blur is very important for multimedia system. Through evaluation and assessment of image blur, the quality of the image processing system can be monitored, and then the system performance can be improved.This paper systematically and intensively studies the no-reference objective blur metric. Firstly, a brief introduction of existing blur assessment algorithms is presented. Then two new algorithms for different application domains are proposed. Specifically for "auto-focus" cases, empirical reveals that the image's histogram concentration to the mean gray level can denote its blur. Accordingly, a new blur metric based on the degree of histogram concentration (i.e. the Hist algorithm) is proposed. The experimental results show that the Hist algorithm can be well applied to assessment of various types of blur, and remains strict monotonicity in strong noise conditions. The Hist algorithm has low computational complexity and can be applied to practical applications. To increase the sensitivity of Hist, two improved methods are proposed: non-linear weighting algorithm (Hist2) and sub-block based algorithm (Histblock). It has been proved that the proposed two methods is effective in improvement of the sensitivity of Hist while still maintaining strict monotonicity.To assess blur in images with different content and blur types, a new blur assessment method based on the width of the step edge is proposed. This method exploits the fact that the width of the step edge diffusion can represent the image blur level objectively. The central idea of this method is to find the features of step edges on a scanning line in a blurred image, and then identifies and calculates the width of the step edge as accurately as possible. Experimental results show that the proposed algorithm can assess out-of-focus blur, Gaussian blur accurately. As for the motion blur, the algorithm is also valid in assessment of the motion distance when the direction of blur is available.