21.02.2013 Views

Advances in Fingerprint Technology.pdf

Advances in Fingerprint Technology.pdf

Advances in Fingerprint Technology.pdf

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

The rationale for determ<strong>in</strong><strong>in</strong>g a s<strong>in</strong>gle orientation for each block of w ×<br />

w pixels (rather than for each pixel) is computational efficiency. Consequently,<br />

<strong>in</strong> regions of a f<strong>in</strong>gerpr<strong>in</strong>t with smoothly flow<strong>in</strong>g parallel ridges,<br />

represent<strong>in</strong>g a s<strong>in</strong>gle ridge orientation for an entire block is not unreasonable,<br />

but <strong>in</strong> the regions where the ridges are sharply chang<strong>in</strong>g their directions (e.g.,<br />

regions surround<strong>in</strong>g core or delta) or the regions with cuts/scars, the choice<br />

of local ridge direction per block is ambiguous. Note that <strong>in</strong> a f<strong>in</strong>gerpr<strong>in</strong>t<br />

image, the ridges oriented at 0° and the ridges at 180° <strong>in</strong> a local neighborhood<br />

cannot be differentiated from each other.<br />

Segmentation<br />

The objective of this stage is to locate the actual region <strong>in</strong> the f<strong>in</strong>gerpr<strong>in</strong>t<br />

image depict<strong>in</strong>g the f<strong>in</strong>ger (region of <strong>in</strong>terest) and discard the regions of the<br />

image conta<strong>in</strong><strong>in</strong>g irrelevant <strong>in</strong>formation (e.g., dirt, smudges leftover from<br />

previous acquisitions, and spurious [pencil] mark<strong>in</strong>gs <strong>in</strong> <strong>in</strong>ked impressions).<br />

This stage is also sometimes referred to as foreground/background detection.<br />

Note that this stage is not responsible for discrim<strong>in</strong>at<strong>in</strong>g the ridges aga<strong>in</strong>st<br />

valleys. A typical approach to segmentation might <strong>in</strong>volve smear<strong>in</strong>g (spatial<br />

gray-scale smooth<strong>in</strong>g) the f<strong>in</strong>gerpr<strong>in</strong>t image and us<strong>in</strong>g fixed/adaptive threshold<strong>in</strong>g<br />

to discard background region. This approach can produce reasonable<br />

results for a good-quality pr<strong>in</strong>t but may not easily remove the extraneous<br />

artifacts <strong>in</strong> a poor-quality f<strong>in</strong>gerpr<strong>in</strong>t image. A method of segmentation based<br />

on the concept of certa<strong>in</strong>ty level of orientation field estimation is described<br />

here.<br />

After the orientation field of an <strong>in</strong>put f<strong>in</strong>gerpr<strong>in</strong>t image is estimated, a<br />

region of <strong>in</strong>terest localization algorithm, which is based on the local certa<strong>in</strong>ty<br />

level of the orientation field, is used to locate the region of <strong>in</strong>terest with<strong>in</strong><br />

the <strong>in</strong>put image. The certa<strong>in</strong>ty level of the orientation field <strong>in</strong> a block quantifies<br />

the extent to which the pixel gradient orientations agree with the block<br />

gradient orientation. For each block, if the certa<strong>in</strong>ty level of the orientation<br />

field is below a threshold, then all the pixels <strong>in</strong> this block are marked as<br />

background pixels. Because the computation of certa<strong>in</strong>ty level is a by-product<br />

of the local ridge orientation estimation, it is a computationally efficient<br />

approach. The authors have found that this method of segmentation performs<br />

reasonably well <strong>in</strong> detect<strong>in</strong>g the region of <strong>in</strong>terest.<br />

Ridge Detection<br />

As alluded earlier, the objective of the ridge detection algorithm is to separate<br />

ridges from valleys <strong>in</strong> a given f<strong>in</strong>gerpr<strong>in</strong>t image. Previous approaches to ridge<br />

detection have used either global or adaptive threshold<strong>in</strong>g, that is, pixels<br />

darker/brighter than a constant/variable threshold are determ<strong>in</strong>ed to be pixels

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

Saved successfully!

Ooh no, something went wrong!