Image processing method for eyes location based on segmentation texture

摘要

Image processing is a very important way of sensing environment. In a facial image, to detect eyes localization should be significant for image understanding and recognition. In this paper, differing from most methods based on color, edge and shape, we present a new method based on segmentation texture, which only depend on features of partial and whole textures. As result of it, the method puts up great robustness in different light, noise, tilted head direction and low cost of calculation. Firstly, a facial image is divided into several square blocks at vertical and horizontal orientation. By employing DFFT (Discrete Fast Fourier Transform), texture of every block is calculated. Textures of these blocks make up segmentation texture, and then some textures features are evaluated. Subsequently, the Hough Transform is involved to detect the facial region in a low-resolution image (27 x 25). Finally, within facial region, the horizontal projection of texture features is computed to locate eye vertical position, and then, to analyze textures of row eye located, the horizontal position of eyes will be obtained. Experiments demonstrate the effectiveness and reliability of this method. (c) 2007 Elsevier B.V. All rights reserved.