Facial feature extraction for head tilt images based on eye canthus

This paper addresses the problem of extracting facial parts such as eyes, nose and mouth from an inclined face. Face image is initially detected using the Viola-Jones algorithm and the facial parts are then cropped and used for many research applications like facial expression recognition, gender classification, identification systems, biometric science, human computer interaction (HCI), surveillance etc. The feature points detected and tracked for a tilted face is difficult using Viola Jones as they depend on rectangles. This work suggests an idea to find the inclined angle of the resultant face with respect to either left or right eye. At first, the eye canthus points are located using Harris corner points and the angle drawn between these localized points with reference to straight line is identified as inclined angle. The input image is then rotated along the inclined angle and the facial parts are detected effectively.