A One Bit Facial Asymmetry Code (FAC) in Fourier Domain for Human Recognition

摘要

The present paper introduces a novel set of biometrics based on facial asymmetry measures in the frequency domain using a compact one-bit representation. A simplistic Hamming distance-type classifier is proposed as a means for matching bit patterns for identification purposes which is more efficient than PCA-based classifiers from storage and computation point of view, and produces equivalent results. A comparison with spatial intensity-based asymmetry measures suggests that our proposed measures are more robust to intra-personal distortions with a misclassification rate of only 4.24% on the standard facial expression database (Cohn-Kanade) consisting of 55 individuals. In addition, a rigorous statistical analysis of the matching algorithm is presented. The role of asymmetry of different face parts (e.g., eyes, mouth, nose) is investigated to determine which regions provide the maximum discrimination among individuals under different expressions.