An efficient algorithm for facial landmark detection using haar-like features coupled with corner detection following anthropometric constraints

Our internal behavior, various situations we are going through, our feelings, all are reflected by face. Facial landmarking is an important step in facial emotions recognition. Landmarking is the process of detection, extraction and localization of fiducial points of face such as silent points on corners of eyes, nose tip, etc. In this paper, we have presented a new automatic facial landmarks detection system which does not require any complicated pre-processing and simple to implement. Our landmarking technique combines Viola-Jones detection algorithm for feature detection with Harris corner detection and then coarse to fine strategy is implemented using an efficient algorithm. Using the Haar like features reduces the cost of brute force search, also provides advantage of speed. Additional selection of sub-regions is also exploited using anthropometric constraints, to limit the search region. This further reduces false detection rate and improves accuracy significantly. A sub-algorithm named Iterative best fit algorithm is used find a land mark exploiting its commonality and geometric configuration and can be used in other contexts as well. The method is tested on JAFEE database and accuracy rate is 93%.