Improving the pose invariant face recognition using skin color and moment information (パターン認識・メディア理解)

概要

論文の詳細を見る
Previously, we proposed a method that implements double frequency analysis (i.e. wavelet and DCT transforms) and color information for extracting global facial features in order to improve the performance of the our previous researches. However, its improvement just improves about 5% recognition rate in rank 1. In order to achieve more improvement we have to consider more information of face images. One of the information that has to be considered is moment information, which provides invariant measure of shape of face images. The moment information can be extracted using normalized central moment. Finally, the global facial features contain not only dominant frequency contents but also the moment coefficients, which provide discriminant information in term of face orientation and face shape. We suppose by implementing the proposed facial features the recognition rate performance will be better that that of the previous one. The experimental results show that recognition rate increase by about 1.5% of without moment facial features with smaller equal error rate (EER).
2008-12-11