A no-reference blur metric guided fusion technique for multi-focus images

This paper presents a new partition fusion technique for multi-focus images based on a no-reference blur metric. The no-reference blur metric is used to evaluate the amount of blur in images whereby, in many practical applications the original images are not available. The proposed algorithm is considered as a region based fusion technique, in which each of the input images are divided into a set of blocks and then the no-reference blur metric is used to evaluate the amount of blur in each block. Finally the less blurred block (sub-image) is selected as a part of the output fused image. In this paper the input images are considered registered whereas; our paper put a focus on image fusion field. The proposed technique considers the presence of overlapped blurred regions in input images; however previous related works didn't consider this assumption. Experimental results show that image fusion based on the selected no-reference blur metric give better performance against other algorithms based on traditional blur measurements.