ROI Extraction of Village Targets and Heterogeneous Image Registration

Since the signal noise ratio of synthetic aperture radar (SAR) image is low as well as speckle noise is much, it's quite difficult to match SAR image with optical image. A new method of target extraction and heterogeneous image matching is proposed in this paper. Median filter and partial differential method are firstly used to remove speckle noise. Next, the GLCM of whole image is calculated by four texture parameters to judge target or non-target points, and a new method which is called local gray-level congruency is proposed to clear false alarm points and increase target points at the same time. Then proper numbers of Fourier descriptor are used to describe ROI for matching and the ROI is filled as closed areas and is marked for final matching. Finally, Hu moment invariants, combined with Euclidean distance and cosine similarity are utilized to complete heterogonous image registration. The experimental results show that the RMSE of final registration is small, the heterogeneous registration image is accurate judged by sight and the whole method owns high reliability, stability and broad application prospect as well.