Resrearch On No-Reference Quality Assessment Of Blur Image

Multimedia is very popular in entertainment and telecommunication. This improves technology of compression and transmission of communication. Whatever, various distortion comes along with processing and communication in every kind of platforms. Blurriness is one of the most common distortions in daily life. Since the blur is caused for missing the high frequency content information in images, it amounts to filter an image by a low-pass filter. It can be observed that if we use different low-pass filters to cover the largest possible range of blur levels, it’s difficult to perceive differences between a blurred image and the same re-blurred image. On the contrary, it’s very easy to discriminate a sharp image and the same blurred one although we use the same low-pass filter. Chapter II in this article propose an improved method, combing edge blocks detection and estimate blur degree in each edge block, to obtain the final blurriness of images.Traditional blurred image quality assessment methods only emphasize the sharpness of edge. These methods’are easy but not accurate enough. So, for recent years, researchers pay more and more attention on the application of HVS on image quality assessment, but few focus on the contribution of saliency. Chapter III in this article proposes a novel no-reference method for image quality assessment based on saliency and SVD. Firstly, for input image, we calculate the initial saliency map and local contrast map. What’s more, a slide window for each pixel is needed to calculate the values of SVD, which is the criterion for each pixel to be sharp or blur. Secondly, combining the local contrast, saliency rectangle RSal and final saliency map Sal is obtained by extending the initial20*20rectangle, from the weighted center of initial saliency map, through some cease criterion. Lastly, we get the image quality IQ by averaging the top10%from the fusion of SVD value, and saliency map Sal. The metric emphases the salient region and remove redundant information, to the greatest extent, to improve the performance of proposed method.