Face Recognition Using Both Geometric Features and PCA/LDA
Youngjun Song, Young-Gil Kim, Nam Kim, Jae-Hyeong Ahn · 2007
To upgrade the performance of the face recognition system, we propose a complex method using the facial geometric feature and principle component analysis/linear discriminant analysis (PCA/LDA). The existing PCA/LDA method can not reflect the facial contour line exactly because it measures similarity according to the degree of physical dispersion. In order to overcome this defect, we measured the distance between the eyes and mouth, and then used this as a feature vector for the energy of each domain within a face, such as the eyes, ears, and chin, to recalculate the similarity originally calculated by the existing PCA/LDA technique when the test and training images differ widely. An evaluation of 400 ORL database face images by using the proposed method confirmed that the recognition rate was increased by approximately 4% in the proposed method compared to that in the existing PCA/LDA method.