High-Resolution SAR Typical Targets Extraction and Heterogeneous Image Registration

Abstract

ABSTRACT To overcome current difficulties in target extraction from high resolution Synthetic Aperture Radar (SAR) and in heterogeneous image registration, new extraction and registration methods are proposed in this paper according to two typical targets: airport and road targets. For airport targets, Otsu segmentation method is firstly used to obtain potential Region of Interest (ROI). Secondly, Fuzzy C-means Clustering Method (FCM) and HU Invariant Moments are jointed to describe regional features of ROI. Finally, Euclidean distance similarity and nearest neighbor distance ratio (NNDR) are jointly used to complete image registration (between optical and SAR image); for road targets, self-adaptive threshold segmentation method is firstly used to obtain ROI. Secondly, the ratio of average (ROA) and Hough transformation are jointed to get the directions of ROI. Thirdly, opening operators using linear operators based on these directions are used to obtain better ROI. Finally, a new method called Triangle Mesh Angle Method (TMAM) is proposed to complete image registration. Experimental results demonstrate the proposed methods have good matching accuracy and robust registration results, and own research value and promising applications.