15.8 Further Reading

Digital Image Processing includes both a discussion of normalized correlation (pp. 583-586) and the logarithmic transform used to compute the output pixels in our sample program (pp. 168-169).

Gonzalez, Rafael C., and Richard E. Woods. Digital image processing. Addison-Wesley, Reading, MA, 1992.

www(imageprocessingplace.com/root_files_V3/publications.htm

J.P. Lewis has an excellent discussion, including a more asymptotically efficient way to accelerate the type of correlation operation implemented by our sample program, where a template match against every pixel in the input image is desired. Lewis uses FFTs to compute the numerators and summed area tables to compute the denominators of the coefficients.

Lewis, J.P. Fast template matching. Vision Interface 10, 1995, pp. 120-123. An expanded version entitled "Fast Normalized Correlation" may be found online at http://bit.ly/NJnZPI.

This page intentionally left blank

The CUDA Handbook Library

As mentioned in Chapter 1, the source code accompanying this book is open source under the two-paragraph BSD license. A pointer to the source code is available on www.cudahandbook.com, and developers can find the Git repository at https://github.com/ArchaeaSoftware/cudahandbook.

This Appendix briefly describes the features of the CUDA Handbook Library (chLib), a set of portable header files located in the chLib/ subdirectory of the source code project. chLib is not intended to be reused in production software. It provides the minimum functionality, in the smallest possible amount of source code, needed to illustrate the concepts covered in this book. chLib is portable to all target operating systems for CUDA, so it often must expose support for the intersection of those operating systems' features.