DT- CWT sub-band partitioning for face recognition

Principal Component analysis extracts the features from the low frequency content of the face image and performs the face recognition with minimal reconstruction error. This statistical method de emphasizes the high frequency information, available to improve the recognition performance. In this paper, the face image is partitioned in to different frequency sub bands prior to PCA analysis. This prior partitioning of the face image, results in more information available for improving the performance of the PCA. Motivated by the shift invariance and Directional property of the Dual Tree Complex Wavelet transform (DT-CWT), this technique is the choice for partitioning the face images. First, the image is partitioned in to different frequency sub bands using DT-CWT, the partitioned sub bands are arranged as a column vector from low frequency to high frequency to form a novel OneS representation. Further, PCA analysis is performed. The resultant feature vectors are classified using k nearest neighbor classifier with Maholanobis cosine distance. Simulations are performed in Matlab, on ORL Database. The DTCWT partitioning of face images is compared with other partitioning techniques like, Discrete Wavelets `db2´ & `coif2´, with baseline as PCA. To show the efficacy of the partitioning, prior to dimensionality reduction, Detection error tradeoff (DET) curves are computed. The DET curves obtained shows that prior partitioning of the face images improves the performance of the further PCA dimensionality reduction.