Extraction of facial feature points using cumulative distribution function by varying single threshold group

This paper proposes a novel adaptive technique to extract facial feature points automatically such as eyes corners, nostrils, nose tip, and mouth corners in frontal view faces, which are based on cumulative distribution function approach by varying different threshold values. At first, the method adopts the Viola-Jones face detector to detect the location of face and also crops the face region with forehead and without forehead areas in an image. The cumulative distribution function of the cropped face region without forehead area is computed first by varying different threshold values to create a new filtered face image in an adaptive way. According to concept of human face structure, the four relevant regions such as right eye, left eye, nose, and mouth areas are cropped from a filtered face image. The connected component of interested area for each relevant cropped filtered image is indicated as our respective feature region. A simple linear search algorithm for eyes and mouth filtered image and contour algorithm for nose filtered image are applied to extract our desired corner points automatically. The method was tested on a large BioID frontal face database with different illuminations, expressions and lighting conditions and the experimental results have achieved an average success rate of 92.89%.