A Stable Region-based Multiscale Image Fusion Scheme with Thermal and Visible Image Application for Mis-Registration Problem

In this paper, we describe a multiscale image fusion method based on shift invariant wavelet transform for decrease misregistration problem of thermal and visible images. The shift invariant wavelet transform avoids the downsampling process. It leads to translation invariant which is available to image fusion. Additionally, the fuzzy possibilistic c-means clustering (FPCM) is applied to perform the multiscale segmentation of source images. The multiscale region representations are obtained and then used to guide the subsequent fusion process. In our experiments, the thermal and visible images are used to be the source images and are the real data. So preprocessing procedure of them is approached here. It includes enhancement and registration based on object matching. Experimental results show that the proposed fusion scheme provides more robust and stable when misregistration distortion exists for the source images