Harris Classic Algorithm, Harris-Laplace Algorithm, and SUSAN Algorithm for Robust Corner Detection

In this document, we explore several image processing algorithms including the Harris classic algorithm, Harris-Laplace algorithm, and SUSAN algorithm. These algorithms are renowned for their exceptional robustness in corner detection applications. The Harris classic algorithm is a widely-used corner detection method that efficiently identifies corner points in images by analyzing the autocorrelation matrix and computing the corner response function R = det(M) - k(trace(M))², where M represents the structure tensor matrix.

The Harris-Laplace algorithm extends the basic Harris approach to multi-scale corner detection, employing scale-space theory to detect corners at different image scales. This algorithm typically implements a scale-adaptive approach by computing Harris corners across multiple Gaussian pyramid levels and selecting stable features using Laplacian-of-Gaussian (LoG) scale selection.

The SUSAN (Smallest Univalue Segment Assimilating Nucleus) algorithm operates by analyzing pixel neighborhoods using a circular mask and comparing intensity values against a central nucleus. Its robustness in noisy environments stems from its nucleus similarity approach, where corner detection is based on the size of the USAN (Univalue Segment Assimilating Nucleus) area rather than gradient calculations, making it less sensitive to noise. These algorithms find extensive applications in computer vision and image processing, each offering distinct advantages for specific scenarios and requirements.