SAR Images Co-registration Based on Gradient Descent Optimization

Abstract

The target of the registration process is to get the disagreement between two captured images for the same area to candidate the transformation matrix that is used to map the points in one image to its congruent in the other image for the same area. A dynamic method is demonstrated in this paper to improve registration process of SAR images. At first, smoothing filtering is used for noise reduction based on gaussian-kernel filter to set aside the pursue-up amplification of noise. Then; area based matching method, cross correlation, is used to perform a coarse registration. The output of the coarse registration is directly applied to the regular step gradient descent (RSGD) optimizer as a fine registration process. The performance of the demonstrated method was evaluated via comparison with the common used corner detectors (Harris, Minimum Eigenvalues, and FAST). Mean square error (MSE) and peak signal-to-noise ratio (PSNR) are the main factors for the comparison. The results show that the demonstrated approach preserves the robustness of the registration process and minimizes the image noise.