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

Utilizing gray-scale integral projection for direct face detection and eye localization is a common algorithm, but its direct application can be affected by background noise and feature variations, resulting in lower recognition accuracy. We propose a detection and localization algorithm combining Otsu's thresholding method (maximum inter-class variance) and region dilation. The algorithm first calculates the Otsu threshold to binarize the gray-scale image and detect face regions, then performs dilation and connectivity processing on connected components within the face area to precisely locate eye coordinates. Experimental results demonstrate that this algorithm is reliable with improved recognition performance. Key implementation aspects include adaptive threshold calculation using Otsu's method and morphological operations for feature enhancement.