Performance Comparison of Two Phase Face Recognition Algorithms based in Frequency Domain
Archana Harsing Sable, Girish V. Chowdhary · 2014
Today, the research in face recognition has focused on developing algorithms that have been proposed to recognize faces beyond variations in viewpoint, illumination, pose and expression. In this paper we have proposed face recognition algorithms, which works in two phase frequency domain i.e., TPFR-DCT-Mah, TPFR-DFT-Mah and TPFR-DWT-Mah. For TPFR-DWT-Mah, the low-frequency subband coefficients i.e. LL subbands (after two-level wavelet decomposition) are used as DWT coefficients, TPFR-DCT-Mah uses the absolute values of DCT coefficients and TPFR-DFT-Mah uses DFT amplitude spectrums to represent the face image, i.e. the transformed image. The first step of the proposed algorithm seeks to represent the test sample as a transformed image and exploits the Mahalanobis distance of each training sample with the test sample to determine K "nearest neighbors" for the test sample. The second step represents the test sample as a linear combination of the determined K nearest neighbors and uses the representation result to perform classification. And finally classify the test sample into the class that has minimum deviation. The proposed algorithm assumes that the K "nearest neighbors" are from the same class as the test sample. The accuracy of the proposed algorithms i.e., TPFR-DCT-Mah, TPFR-DFT-Mah and TPFR-DWT-Mah has been identified and a comparison was performed between them in terms of recognition rates or Equal Error Rate(EER) or the Receiver Operating Curves(ROC).