Gray-scale Integral Projection for Direct Face Detection and Eye Localization

In face image detection and eye localization, direct processing using gray-scale integral projection is a commonly employed algorithm. However, direct implementation of this algorithm is susceptible to interference from background elements and feature variations, potentially leading to reduced recognition accuracy. To address this limitation, we propose a novel detection and localization algorithm based on the combination of Otsu's thresholding method (maximum inter-class variance) and region dilation. Specifically, the algorithm first computes the Otsu threshold to determine an optimal binarization point (implemented through histogram analysis and variance maximization between foreground/background classes), then converts the gray-scale image into a binary image to detect face regions. For precise eye coordinate localization, we perform dilation operations (using structural elements like disks or squares) and connectivity processing on connected components within the detected face area. Experimental results confirm that our proposed algorithm is reliable and achieves improved recognition performance. Consequently, this method can be effectively applied to face detection and eye localization tasks with broad application prospects. The implementation workflow involves image binarization, connected component analysis, and morphological processing for robust feature extraction.