A robust personal identification through faces based on isodensity maps

Abstract

An efficient personal identification system which is adaptable for either rotated or tilted faces is presented. In the proposed method, the 25 facial images, taken with different angles, and their compressed images, reduced to a quarter and one-sixteenth in size, are registered in the database as reference images. The similarity of an input image is first examined between all registered compressed images, and the three facial images, which are similar to the input image, are then extracted as the candidate facial images. Isodensity maps, expected to have maximum similarity to the input image, are created by interpolation between two candidate images. Finally, the input image is judged identical or not using the candidate and interpolated images. Furthermore, the effectiveness of removing spectacles from input images is discussed. Computer simulation with 50 people shows 95.21% correct discrimination accuracy. In addition, the processing time is reduced to approximately one-thirtieth. From the experimental results, the prospects of using this method are very encouraging.