Abstract
In the present paper, a new face model for detecting human faces in still images was developed by observing the relative geometric positions of facial features and the variations in the pixel values. These observations were converted into a set of rules which were mathematically expressed. To evaluate the performance and identify the facial characteristics in the proposed method, an algorithm was implemented and evaluated on a test set of 72 images containing 137 faces. The evaluation was based on the number of faces correctly detected and the correctness of the feature positions. The detection process was controlled by parameters that could be adjusted by users, at will, based on the characteristics of input images. The proposed method is simple, flexible, accurate and efficient. Owing to its simplicity, it can be easily implemented in most existing surveillance and automation systems.