20.01.2013 Views

Master Thesis - Department of Computer Science

Master Thesis - Department of Computer Science

Master Thesis - Department of Computer Science

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Image<br />

Rows<br />

h(.)<br />

g(.)<br />

2<br />

2<br />

Columns<br />

Figure 3.2: Two dimensional discrete wavelet transform (DWT) for an image: The<br />

low pass filter h(.) and the high pass filter g(.) are applied along the rows initially<br />

followed by downsampling by a factor <strong>of</strong> two. This is followed by filtering using h(.)<br />

and g(.) along the columns and downsampled by factor <strong>of</strong> two.<br />

<strong>of</strong> the wavelet decomposition can be further decomposed using the DWT, giving a<br />

multi-level dyadic subband decomposition <strong>of</strong> the face. In the following subsection,<br />

the method <strong>of</strong> reconstructing a subband face using selective wavelet-decomposed sub-<br />

bands is discussed.<br />

3.2.2 Subband Face Reconstruction<br />

A face image <strong>of</strong> a person contains common (approximations) as well as discriminatory<br />

(details) information with respect to faces <strong>of</strong> all other persons. The discriminatory<br />

information is due to structural variations <strong>of</strong> the face which are acquired as intensity<br />

variations at different locations <strong>of</strong> the face. The location and degree <strong>of</strong> intensity vari-<br />

ations in a face <strong>of</strong> an individual are unique features which discriminate one from the<br />

rest <strong>of</strong> the population. The similarity <strong>of</strong> a face with respect to another is in the global<br />

appearance and structure <strong>of</strong> the face. These information (similar and discriminatory)<br />

are segregated at different subbands using different levels <strong>of</strong> decomposition <strong>of</strong> the<br />

face image. Wavelet decomposition helps to split the features <strong>of</strong> a face in different<br />

subbands with “approximations” containing the common (smooth) parts <strong>of</strong> the face<br />

and “details”, at certain levels <strong>of</strong> decomposition, containing the discriminatory (vari-<br />

49<br />

h(.)<br />

g(.)<br />

h(.)<br />

g(.)<br />

2<br />

2<br />

2<br />

2<br />

LL<br />

LH<br />

HL<br />

HH

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!