Fractional-order differentiation based sparse representation for multi-focus image fusion

Abstract

The aim of image fusion is to obtain a clear image by combining useful information coming from multiple images, so it is crucial to extract the salient features of source images as the activity level measurement effectively. In this paper, a novel algorithm called fractional-order differentiation based sparse representation (FD-SR) is presented for multi-focus image fusion. In this algorithm, the source images are first convoluted with fractional-order differentiation masks to acquire the feature maps, from which the histograms of oriented gradients (HOG) are computed to capture human vision-related salient information. Next, to construct a representative dictionary for sparse representation, the HOG patterns are then partitioned into many patches which are clustered to retain the structural information. From these clusters, compact sub-dictionaries are learned using orthogonal matching pursuit (OMP) respectively and then combined to form the overcomplete dictionary. Finally, the fused sub-images are reconstructed with the dictionary based on max l1 rule, and all these sub-images constitute the whole fused image. The experimental results on multi-focus image datasets and medical image dataset validate the effectiveness of the proposed method for image fusion task.