Difference of Gaussian (DoG) edge detection added

[wang1211]

Another edge detection method besides sobel. DoG can approximate Laplacian of Gaussian operator. It works as an efficient edge detector.

Will add a demo soon.

[coveralls]

Coverage decreased (-0.73%) to 96.4% when pulling 7ddcd9c on wang1211:master into cf3463f on luispedro:master.

[coveralls]

Coverage decreased (-0.73%) to 96.4% when pulling 7ddcd9c on wang1211:master into cf3463f on luispedro:master.

[coveralls]

Coverage decreased (-0.73%) to 96.4% when pulling 7ddcd9c on wang1211:master into cf3463f on luispedro:master.

[luispedro]

Thank you. This looks very good.

I think I can merge as is and add a simple test later or you if you have a test, that'd be nice too (also needed: a credit for you in the ChangeLog file).

Is the sigma2 = sigma1 * 1.001 relationship "standard" (used by other packages &c)? Should it be configurable?

[wang1211]

Thanks for the feedback. From what I have learned, 1.001 is not "standard" but a small multiplier such as 1.001 should make it approximate Laplacian of Gaussian better. However, after looking on the internet, I have also seen people using large multiplier such as 2. So a configurable parameter might be better in case users need it.

[coveralls]

Coverage decreased (-0.73%) to 96.4% when pulling b3058d5 on wang1211:master into cf3463f on luispedro:master.

[coveralls]

Coverage decreased (-0.95%) to 96.18% when pulling be7c509 on wang1211:master into cf3463f on luispedro:master.

[luispedro]

Thank you again. I think this all looks good and I'm merging now.

[wang1211]

Cool. Thanks!