Satellite Imagery Analysis Using Image Registration and Edge Detection

In this research we evaluated multiple feature-based image registration techniques by leveraging keypoint detectors to extract robust and distinctive key points for matching across images. Various detectors and descriptors, including those used in the ratio test, FLANN parameters for approximate nearest neighbor searches, and the BF Matcher for brute-force matching, were employed to extract keypoints, which were then used to search for suitable matches. Various techniques, including warp Affine, estimateAffinePartial2D, and homography, were employed to estimate transformations. We have done a comparative analysis in which we used different combination of feature-based image registration algorithms and edge detection algorithms for the purpose of image alignment and boundary extraction. Subsequently, a performance assessment was performed by comparing the registered images with ground truth using Structural Similarity Index Measure (SSIM), Precision, Recall, Mean Square Error (MSE) and F1-Score metrics. Among the evaluated models, the BRISK model with the Canny edge detection algorithm consistently demonstrated superior performance, achieving the highest Precision (0.973), Recall (0.984), F1-Score (0.918), SSIM (0.0427), and the lowest MSE (0.0427) scores. These findings emphasize the efficacy of the BRISK model in image alignment tasks.