A Hybrid Algorithm of Eyes Localization in Color Facial Region

In this paper, a hybrid algorithm for precise eyes localization in color facial region is presented. In practice, most eye detection suffer from the influence of illumination and face pose. Multiple techniques are integrated in this algorithm to overcome these limitations, such as color space mapping, illumination correction, mouth detection, dynamic threshold and so on. The scheme consists of two stage: eye region extraction and fine eye location. Firstly, with aids of clustering information of eyes in color image and gray-level image, we highlight possible eye regions. Through binarization and morphological operation, we can reserve major blocks and remove most of tiny blocks. For accurate extraction to the best, the threshold of binarization is adjusted appropriately according to heuristic rules test feedback. Heuristic rules are established in terms of priori knowledge about geometrical relationship among facial components. Thereby, we can filter out most of non-eye candidates and allow more possible candidates to be selected as eyes. In the stage of fine eye location, we employ twice extraction to find eye center via projection function and verify validation of location by similarity matching finally. With regard to variable lighting, we have adopted an effective local illumination correction to reduce interference. Some other experiments are carried out to investigate method capability on variations of pose, facial expressions, partial occlusions and lighting conditions on face images. Experiment results indicate our algorithm can achieves promising performance under certain illumination and various face poses conditions.