Study In Multi-focus Image Fusion Based On Region Segmentation And NSCT

As an important branch of information fusion, Multi-focus image fusion mainlysolve the practical issue that for the optical lens’s depth of field, all objects can’t getclear image at the same time in different focal plane but same scene. And Multi-focusimage fusion can get an image higher resolution and greater confidence by the used ofthe redundancy and related information between some multiple images of differentfocus points.This paper mainly done the following aspects of research which is based on theresearch and analysis of the imaging characteristics of Multi-focus image, togetherwith the existing algorithm and effect evaluation of the fused multi-focus image.(1) Theory analysed and experiments proved that the fusion effect affecting bythe decomposition levels of the multi-scale decomposition by NonsubsampledContourlet Transform (NSCT).So we can get a level which obtain better objectiveevaluation as good support for parameter selection in future.(2) According to the characteristics of multi-focus image and NSCT, amulti-focus image fusion algorithm which based on segmentation information andNSCT was advanced. The algorithm done the edge information detection for sourceimage by mathematical morphology with multi-scale and multi-structuring elements,then can estimate the definition by the Edge information’s area Energy. From thedefinition,the Source image can be divided into clear, blurry,and transition regions.The coefficient of clear area is still adopted as fusion coefficient, as to the area fromclear to blurry, the low-frequency coefficients obtained base on weighted by edgeenergy of the source image, but the high-frequency coefficients got by the method ofthe biggest absolute value. In addition, the algorithm was proved by comparingexperiments.(3)Regard to noised images and low-noised images, two multi-focus imagefusion algorithms with improved fusion rule are advanced based on the differentcharacteristics of the frequency coefficients under NSCT. For low noise images,Low-frequency coefficients depend on both.the edge information of low frequencysub image and the regional energy of High frequency coefficients. But forhigh-frequency coefficients, the highest level high-frequency coefficients and otherhigh-frequency coefficients both have own character which result in that the formerfigured by the biggest absolute value but the latter adopt the regional energy method.Besides, the low frequency has less influence by the noise and the most of noisesexists in the highest levels high frequency coefficients,so we can use the methodcombined with edge information of low frequency sub image and window’sinformation for the highest levels high-frequency coefficients,as to other coefficients,we can take the region Acutance of the low frequency sub image. In some extent,the method mentioned abovecan reduce noise.