Master Thesis - Department of Computer Science
Master Thesis - Department of Computer Science
Master Thesis - Department of Computer Science
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
CHAPTER 3<br />
An Efficient Method <strong>of</strong> Face Recognition<br />
using Subject-Specific Subband Faces<br />
This chapter presents an efficient method for frontal face recognition, using subject-<br />
specific subband face representation. The human face has certain visual features that<br />
are common among everybody and some others that exhibit the unique characteris-<br />
tics <strong>of</strong> an individual. Using the discrete wavelet transform (DWT), we extract these<br />
unique features from the face image for discriminating it from others. The face image<br />
is decomposed into several subbands to separate the common (approximation) and<br />
discriminatory (detail) parts. Subband face is reconstructed from selected wavelet<br />
subbands, in which a suitable approximation subband is suppressed and a detail sub-<br />
band (in some cases) is eliminated. Reconstructing a face with an optimal selection<br />
<strong>of</strong> subbands enhances the performance <strong>of</strong> face recognition. We present four different<br />
criteria as cost functions to obtain an optimal subband face for each subject, and<br />
compare their performances. The performance <strong>of</strong> the subband face representation on<br />
several linear subspace techniques: PCA, LDA, 2D-PCA, 2D-LDA and Discriminative<br />
Common Vectors (DCV) with Yale, ORL and PIE face databases shows that the sub-<br />
band face representation performs significantly better than that proposed by Ekenel<br />
for multiresolution face recognition [36] for frontal face recognition, in the presence<br />
<strong>of</strong> varying illumination, expression and pose. The rest <strong>of</strong> the chapter is organized as<br />
follows. Section 3.1 provides the motivation and a brief introduction to our proposed<br />
method along with a concise overview on wavelet based face recognition techniques.<br />
The Wavelet Decomposition and the method to reconstruct a subband face is de-<br />
scribed in section 3.2. Section 3.3 highlights over the criteria used as cost functions<br />
and also the algorithm to select optimum subject-specific subbands. Section 3.4 dis-<br />
cusses the experimental results <strong>of</strong> the proposed subband face representation tested