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.
improve the accuracy further. They obtained a maximum accuracy <strong>of</strong> 96.67% using<br />
ICA2 on a database (for expression variations) comprised <strong>of</strong> 272 images from CMU<br />
PIE and rest 328 from FERET. For finding subbands insensitive to illumination they<br />
experimented on a database consisting <strong>of</strong> 272 images from CMU PIE and remaining 60<br />
images from Yale database and reported a maximum accuracy <strong>of</strong> 77.71%. The main<br />
drawback <strong>of</strong> this method lies in the adhoc method <strong>of</strong> selecting successful subbands<br />
on a testing set which involves over-tuning to attain maximum possible accuracy.<br />
None <strong>of</strong> these approaches have attempted to exploit the fact that, if some <strong>of</strong><br />
the common visual features <strong>of</strong> a face exist in a certain subband and are suppressed<br />
during reconstruction, we obtain a subband face with only the discriminatory infor-<br />
mation. The retina <strong>of</strong> the human visual system has been observed to perform local<br />
multi-resolution, multi-channel processing <strong>of</strong> the signal using a bank <strong>of</strong> tuned band-<br />
pass filters [65, 85, 98]. The visual cortex uses second order relational differences in<br />
structure with respect to an average face for perception [41, 56]. Hence, a unified<br />
computational model consisting <strong>of</strong> DWT/IDWT (signal processing) and PCA/LDA<br />
(statistical processing) will be able to closely imitate the human visual system, more<br />
effectively than what statistical processing does alone. These studies motivated us to<br />
explore the use <strong>of</strong> IDWT to obtain a subband face, by suppressing an approximation<br />
and retaining some <strong>of</strong> the discriminating subbands in the process <strong>of</strong> image reconstruc-<br />
tion, and then use the subband face for recognition using statistical approaches.<br />
In this chapter, we propose the subband face as a new representation for the<br />
face recognition task. Only the discriminatory information <strong>of</strong> a face is retained or<br />
captured in a subband face. Discrete wavelet transform is used to decompose the<br />
original face image into approximation and detail subbands. We perform multi-level<br />
dyadic decomposition <strong>of</strong> a face image using the Daubechies filters [30]. The subband<br />
face may be reconstructed from selected subbands by suppressing the approximation<br />
at a suitable higher level and retaining the details. An inherent information fusion<br />
is being performed by the reconstruction process which retains only the inter-class<br />
discriminatory informations and discards the inter-class common informations. The<br />
information <strong>of</strong> a face in the details at lower levels <strong>of</strong> decomposition, usually contains<br />
noise and redundant pixels which do not constitute any discriminatory information<br />
46