04.01.2015 Views

Fusion of Visual and Thermal Face Recognition Techniques: A ...

Fusion of Visual and Thermal Face Recognition Techniques: A ...

Fusion of Visual and Thermal Face Recognition Techniques: A ...

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.

Histogram based automatic thresholds make it even harder to find the optimal<br />

threshold for segmenting facial components. Figure 8 shows the histograms <strong>of</strong> an original<br />

<strong>and</strong> equalized thermal image. Since the optimal threshold, which can be derived from the<br />

histogram, does not contain any regional information, the approach to solve local minima<br />

or local maximum has it own limitations.<br />

(a)<br />

0<br />

(b)<br />

1<br />

(c)<br />

0<br />

(d)<br />

1<br />

Figure 8: Problems with histogram methods (a) original thermal image, (b) histogram <strong>of</strong><br />

(a), (c) equalized thermal image, (d) histogram <strong>of</strong> (b)<br />

Therefore, we make use <strong>of</strong> average thermal images to measure variation over thermal<br />

face images instead <strong>of</strong> observing each individual thermal face. Figure 9 shows the<br />

average <strong>of</strong> thermal images generated from 60 images without eyeglasses (a) <strong>and</strong> with<br />

eyeglasses (b). Figure 9(c) was generated from 200 images from the FERET database.<br />

All images are normalized using manual eye positioning.<br />

25

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

Saved successfully!

Ooh no, something went wrong!