A robust face recognition approach against variant illumination

In order to alleviate the effect of the light illumination and environment noise, a robust face recognition method is proposed in this paper based on Curvelet transform and local ternary pattern. The Curvelet Transform (CT) is a new anisotropie multi-resolution technique, which can effectively retain image edge information. Local Ternary Pattern (LTP) is an extended version of Local Binary Pattern (LBP). First the face images are decomposed into three parts by CT, and then we process the coefficients of its first band by using logarithm computation and LTP, while directly delete the redundant highest frequency information in the third part with an aim of removing the environment noise and the noisy information at the intersection of the light and the object. Then we select the principal features from the second part coefficients by using Principal Component Analysis (PCA). Finally, the face recognition is done by using Linear Discriminant Analysis (LDA) with the preprocessed first part features and the second part features obtained from PCA. Extensive experiments show that the proposed method can alleviate the effect of the illumination and environment noise effectively, which achieves better face recognition rate than the Curvelet+PCA+LDA.