Face detection from color images by iterative thresholding on skin probability maps

In this paper, we present a technique for face detection from color images, which utilizes facial geometry and iterative thresholding of skin probability maps to determine candidate face regions in an image. The actual face regions are determined by a fast and efficient eye locator technique that employs symmetry analysis along with luminance, chrominance and curvature information. Iterative thresholding makes the technique tolerant to variations in illumination and background. The facial geometry analysis and eye locator techniques are tolerant to scale variations of faces. Experimental results demonstrate the superiority of the proposed technique to a standard PCA (principal components analysis)-based technique, in the presence of scale and illumination variations.