A method to reduce the computational cost of Modified Hausdorff Distance in Face Recognition
Tinh T. Bui, Truong Thien Nhan, Dang Nguyen Chau · 2019 6th NAFOSTED Conference on Information and Computer Science (NICS) · 2019
The Hausdorff distance (HD) is a metric to measure the similarity between two sets of points. Modified Hausdorff distance (MHD) method in face recognition is the applying of the Hausdorff distance to find the correlation between an input image and the model images in the database, thereby giving the best-matched image. It gives high accuracy in comparing with prevalent methods. However, the calculated volume in this method is very large, not suitable for high-speed identification systems with huge databases. In this paper, we propose a new method to reduce the computational complexity of MHD in face recognition. Firstly, we vectorized the dominant points (DPs) extracted from the face image based on the correspondence between them. Then, instead of completely scanning all points, our method quickly computes the component distances of MHD using the angles of representative vectors corresponding with DPs. Experimental results indicate that the complexity of the proposed method has improved by 8.41-8.75 times decrease. Besides, the accuracy of the proposed method in face recognition is also compared with the MHD method under varying conditions of the faces: 1) ideal condition, 2) various lighting conditions and 3) different poses of the head. The performance of the proposed method has been higher than the initial method under the ideal condition and various lighting conditions, and as equivalent to the original method under another condition.