S.-i. Murakami and S. Wada (Japan)
Face recognition, Kernel Based Fisher Discriminant Analysis (KFDA), Color space, DAGSVM
In this paper, we would propose a robust face feature extraction method based on KFDA. Using extracted features from color face images, we generate feature vectors and apply them to the personal recognition system. In feature extraction process, a color face image represented by RGB space is transformed to YCbCr space as well as HSV space. The luminance component (Y), the dark color difference chroma component (Cb) and the saturation component (S) are used to extract robust feature vectors. First, the KFDA is applied to each component in order to reduce the influence of image capture conditions such as face size deviation and illumination change. Next, the feature vector generated with those of Y, Cb and S component is applied to discriminate the face by DAGSVM. Computer simulations for face recognition are executed to show the effectiveness of our method. Face images that are degraded by capture conditions are used as a query input images. It is confirmed that our feature extraction method is robust to face size deviation and illumination changes. It is shown that the high face recognition rate is achieved compared with the conventional method using a single component in several actual conditions.
Important Links:
Go Back
