Study and Performance Analysis of Image Fusion Techniques for Multi-focus Images

Abstract

The primary objective behind image fusion technique is to collect and integrate all the essential as well as relevant features and information in a solitary image. In case of multi-focus image fusion technique, the procedure involves accumulation of information out of the focused regions from the input images and final combined outcome (image) will contain all the focused regions as well as objects. There have been several studies in regard to multi-focus image fusion technique in the area of spatial domain as well as transform domain. The issue of appearance of non-focused regions or objects in an image happens due to limited depth of field of the camera lens. As a result, objects present in the focused region of the camera lens appears focused and others appear as un-focused. Image fusion technique is a scheme to overcome this issue. In this paper, authors have reviewed recent fusion-based techniques and tested certain image fusion techniques, i.e., discrete wavelet transform, i.e., DWT, independent component analysis, in short, ICA, sparse representation, i.e., SR, dual-tree complex wavelet transform, i.e., DTCWT, non-subsampled contourlet transform abbreviated as NSCT, and a hybrid of NSCT + SR, on Lytro multi-focus image dataset and comparatively analyzed these methods on fusion metrics nonlinear correlation information entropy (NCIE), normalized mutual information (NMI), gradient-based metric(GBM), phase congruency-based metric (PCB). Analysis has demonstrated that NSCT + SR has given best performance results with a NCIE of 0.842, NMI of 1.121, GBM of 0.759, and PCB of 0.848, while method SR has given second best performance result. Authors have also elucidated regarding the essential requirements to consider while framing any fusion-based scheme.