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.
each subject, 42 samples (flashes from 21 different directions with and without<br />
the room lights on) were used for our study. Henceforth we use the following<br />
notations for each <strong>of</strong> the 21 flashes with and without room lights on:<br />
(i) 21 flashes with room lights on: lights flash02 - lights flash22,<br />
(ii) 21 flashes without room lights on: illum flash02 - illum flash22.<br />
The face samples were extracted from the original gray level images for the<br />
PIE database, based on the specification <strong>of</strong> eye locations given in [5].<br />
• The ORL face database has 40 subjects with 10 samples each. There is no<br />
change in illumination but significant (near frontal, no pr<strong>of</strong>ile views) change in<br />
the face pose.<br />
3.4.2 Performance Analysis on Three Standard Face Databases<br />
To select subject-specific optimal subbands we split the image database into three<br />
disjoint sets, namely training, validation and testing set. Training set along with<br />
validation set are used for subband selection, and then the performance <strong>of</strong> selected<br />
subbands is observed on the testing set. The number <strong>of</strong> images used for a subject<br />
over the three sets, for all three databases are given in Table 3.1. Fig. 3.9 shows<br />
the images in training set for a single subject on Yale, PIE and ORL databases. For<br />
subband selection, the image size has been kept as 64*64 for all subspace methods<br />
except DCV where it is 25*25. The maximum values for l and k are 2 and 4 for DCV,<br />
while for other subspace methods they were chosen as 4 and 7 , respectively. For the<br />
PIE database, which has 42 frontal samples, only 4 face images corresponding to 2<br />
frontal flashes (illum flash08, lights flash08, illum flash11, lights flash11) are used for<br />
training.<br />
Table 3.2 shows the performances <strong>of</strong> subspace methods on original images for<br />
three databases. Table 3.3 shows the verification (in terms <strong>of</strong> Equal Error Rate or<br />
EER) and recognition (in terms <strong>of</strong> Peak Recognition Accuracy or PRA) performance<br />
<strong>of</strong> subband face representation integrated with subspace methods (PCA, LDA, 2D-<br />
PCA, 2D-LDA and DCV) on Yale database. The results in the column labeled<br />
“Subband Face(C1)” corresponds to that obtained using optimal subband faces for<br />
subjects as selected by minimizing the first criterion discussed in Section 3.3.1. Same<br />
61