A New Approach For Multi-source Image Fusion

Multi-sensor image fusion refers to the techniques that integrate complementaryinformation from multi-image sensor data so that the fusion images are more suitablefor the purpose of human visual perception and computer processing. Each type ofsensor is applied in some given application including remote sensing,medical diagosis,computer vision, etc. Moreover, the existing image fusion approaches can be classifiedinto three categories: pixel level, feature level and design level. This paper is focused onthe pixel-level image fusion approach.In this paper, we mainly focus on the methods of image fusion. The structure,purpose, applications of image fusion is involved as well. Firstly,some routine fusionmethods are introduced in detail, including average approach, pyramid method andwavelet method. Then the preprocessing steps of image fusion are described in details,which include the image denoising and image registration. Furthermore, an exactmaximum likelihood (EML) registration method, which combines both control pointand intensity, is proposed for image registration. The experimental results demonstratethat the proposed approach achieved higher registration accuracy.Since the limited depth-of-focus of optical lenses, it is often not possible to get animage which contains all relevant objects in focus. A novel optimal method formulti-focus image fusion using differential evolution algorithm is proposed. Firstly, thesource images are decomposed into predefined size blocks. Then the sharper blocks areselected by employing a sharpness criterion function. The selected blocks are used toconstruct the fused image. The motivation of the proposed approach lies in the fact thatan optimized block size could be more effective than a fixed block size. Moreover, theexperimental results demonstrate that the proposed approach performs better than theother traditional approaches in terms of both visual and quantitative evaluation.