Human Face Recognition
Another important benefit from the use of correlation filters is the resulting automatic shift invariance. Since correlation filters are linear shiftinvariant filters, any translation in the test input image will result in correlation output being shifted by exactly the same amount. Since the first thing we do in processing correlation outputs is to locate the correlation peak and since filters are designed to yield centered correlation peaks for centered training images, we do not need to explicitly center the test image. Instead, we implicitly center the test image by locating the correlation peak and
computing the PSR centered at this peak location. In many feature-based FR methods, centering of a test image prior to computing the feature values is critical since they are
often sensitive to image centering. We illustrate the shiftinvariance property of the correlation filters in Fig. 6. The left half of this figure shows an occluded and off-centered test image and the right half shows the resulting correlation output. The peak is still very visible except that it has moved to a new location. The shift-invariance is not just a theoretical curiosity. In practice, even with our best algorithms, centering test images is only approximate and can be time consuming. Thus, methods that avoid the need for explicit test image centering are attractive.
Want To Know more with
Contact for more learning: webmaster@freehost7com
The contents of this webpage are copyrighted © 2008 www.freehost7.com
All Rights Reserved.