Harris Multi-Scale Corner Detection Algorithm Implementation

We have developed an innovative Harris multi-scale corner detection algorithm designed to significantly improve registration accuracy in image alignment algorithms. This approach incorporates multi-scale image analysis through Gaussian scale-space implementation, where corner response functions are computed across multiple sigma values (typically σ=1.0, 1.5, 2.0) using the standard Harris corner detector formula: R = det(M) - k(trace(M))². The algorithm then performs non-maximum suppression across scales to identify stable corner points that persist through different resolution levels. This multi-scale framework enables more accurate corner localization by overcoming single-scale limitations where corner detection sensitivity varies with image resolution. The implementation involves computing image gradients using Sobel operators, constructing the autocorrelation matrix M for each scale, and integrating scale-adaptive thresholding to distinguish prominent corners from texture noise. Our method provides robust corner detection that enhances feature matching precision in image registration pipelines, with applications spanning computer vision systems, medical image analysis, and remote sensing image processing. The algorithm can be implemented using OpenCV's cv2.cornerHarris() function extended with custom scale-space loops, offering researchers and practitioners an effective solution for high-precision feature-based image alignment tasks.