Research On Image Fusion Based On Multi Scale Transform

Complementarity and redundancy exist between multiple source images. In order to overcome the shortage of providing limited information in a single source image, this paper conducts an investigation into multi-scale geometric transformation, discusses image fusion algorithm based on multi-scale geometric transformation.The main contents and innovations are as follows:1.The application of Smoothlet transformation in image fusion field. Smoothlet transform has good properties of "line" and "side", it has a good approximation for smoothing edge information, and it can effectively extract the edge information of the image those need to be retained. But Smoothlet has a limit on the directions when performing image approximation, so combined Smoothlet with other multi-scale transforms to apply to image fusion.2.The algorithm of combining Smoothlet with NSCT. In the process of image fusion, the activity-level of NSCT transform in the spatial domain of the image is not high, and the information extracted from the source image is incomplete,leads to generate various degree blurs when fused image in the position of details, so it will has a bad effect on the accuracy of image. Aimed at the lack of NSCT transform, this paper presents a fusion algorithm combining NSCT with Smoothlet, NSCT has a good directional property, so the two methods have a good complementarity. And it retains the approximation image and residual image in fusion process, so it can extract the feature information which need to retain in images more effectively. Simulation results show's that the fusion algorithm can welled retain the information detail of the source image.3. The algorithm of combining Smoothlet with NSST. Single multi-scale decomposition method often adapts to a single particular type of images, but not suitable for other types of image. Applied this method that combining Smoothlet with NSST to multi-focus image and medical image fields can receive a good fusion effect. NSST is constructed by "Meyer" wavelets, it can obtain compact support structure of multi-direction. During Shearlet operation, numbers of direction can be set, increased numbers of direction in images makes the fused image more precise. By combining with NSST method, Smoothlet can make up the limitation of the directions and retain the approximation images and residual images, so it can improve the retaining degree of source image information when performing NSST. Simulation results show's that the image effect of this algorithm is better than that traditional single multi-scale geometric method.