Eye Corner Detection Using SUSAN Algorithm on MATLAB Platform

On the MATLAB platform, the SUSAN (Smallest Univalue Segment Assimilating Nucleus) algorithm can be employed for detecting eye corners. The SUSAN algorithm operates as a sub-pixel level corner detection method that precisely locates feature points in images while maintaining strong stability under varying lighting conditions and noise interference. In practical implementation, MATLAB's image processing toolbox facilitates the algorithm through functions like fspecial for creating circular masks and conv2 for correlation operations. The core algorithm compares pixel intensity differences within a circular neighborhood (typically 37 pixels) against a brightness threshold to identify corner regions. For eye detection applications, preprocessing steps may include iris localization using Hough transform (imfindcircles) before applying SUSAN to enhance accuracy in eye tracking and recognition systems. Furthermore, the SUSAN algorithm can be integrated with other tracking methods such as the Kanade-Lucas-Tomasi (KLT) algorithm (vision.PointTracker) for ocular motion tracking, where SUSAN provides initial corner points while KLT handles temporal continuity, resulting in more precise eye movement trajectory analysis.