To search, Click below search items.


All Published Papers Search Service


Regularized Independent Component Analysis Regularization in Face Verification


Pang Ying Han, Ooi Shih Yin, Teo Chuan Chin, Low Cheng Yaw, Hiew Fu San, Goh Kah Ong


Vol. 13  No. 10  pp. 54-60


In this work, a regularized Independent Component Analysis (coined as RICA) is proposed in face verification. RICA attempts to generate a linearly independent feature representation with minimal within-class variance, leading to better data discrimination. In RICA, information of correlation coefficients between image data is employed to form a Laplacian matrix. This matrix measures the degree of deviation of a data point from its nearby/ adjacent points. In other words, local discriminative features of data could be disclosured through the Laplacian matrix. As the name suggests, independent component analysis (ICA) is adopted as feature descriptor in this approach. Since there are two different architectures of ICA (ICA I and ICA II), RICA is implemented on these two types of feature extractor and the proposed techniques are known as RICA_ICA I and RICA_ICA II, respectively. The efficiency of RICA is assessed based on three face datasets, namely (1) Facial Recognition Technology (FERET), (2) CMU Pose, Illumination, and Expression (CMU PIE) and (3) Face Recognition Grand Challenge (FRGC).


Face verification; Correlation Coefficient; Laplacian Matrix; Regularization; Independent Component Analysis