New Filter Based Unsupervised Rules for Boolean Blur Metric

This paper presents a simple and novel no reference filter based rules for Boolean blur metric (FBBM) to classify the blurred images from the database of home photos. Our primary goal of this paper is to classify the blurred images rather than measuring degree of blurness in the image or deblur an image. Thus the name given to this approach is Boolean blur metric (BBM). The proposed approach explores new rules based on establishing the unique relationship between the arithmetic mean filter, geometric mean filter and median filter of given image with the help of canny edge detector. The metric uses the disadvantage of arithmetic mean filter and advantage of geometric mean filter and median filers to define Boolean rule. Further, we have shown that the number of canny edge components in filtered images makes difference in defining rule. Finally, the proposed approach is compared with the well known existing no reference perceptual blur metrics to show that existing metrics are not suitable for classification. In addition, the experimental results revealed that the proposed method works even for rotated and scaled images