An improved LBP algorithm for avatar face recognition

Abdallah A. Mohamed, University of Louisville
Roman V. Yampolskiy, University of Louisville


This paper presents a novel avatar face recognition algorithm based on Discrete Wavelet Transform and LBP descriptor. The 2-D Discrete Wavelet Transform has been used to process the avatar face dataset by extracting the low frequency components and then forming the low frequency sub-images. Then the LBP operator is used to extract the characterizations of these sub-images. Finally, the Chi-Square distance is used to connect each image to its subject. Experimental results show that the proposed method can be used effectively in avatar face recognition with a single training sample per avatar. The performance of the proposed algorithm on a dataset of 581 avatar face images shows that the proposed algorithm performs better than PCA and Traditional (single scale) LBP method with respect to the recognition rate. © 2011 IEEE.