Kovesi's functions

afre · January 27, 2019, 2:14am

This may fit into my never ending G'MIC exercises thread but I have decided to make a new one. Came across Kovesi’s site yesterday which has lots of interesting goodies. His GitHub has examples.

– Peter's Functions for Computer Vision
– Examples · ImagePhaseCongruency
– PerceptualColourMaps.jl/README.md at master · peterkovesi/PerceptualColourMaps.jl · GitHub

In particular, I have been interested in ways to overcome gradient_norm's weakness of doubling thin features such as the whiskers of the tiger and variable feature outline thickness and brightness. Observe:

Comparison using his sample images (left); gradient_norm (centre, gmic), phase congruency (right, older faster version).

Perceptual colour maps are interesting as well. Reminds me of my Palette to clut to application thread among other related discussions.

Reptorian · January 27, 2019, 2:21am

Still on the quest to improve gradient norm? Those colourmap would be interesting to convert to GIMP/Krita compatible gradient map.

afre · January 27, 2019, 2:45am

Isn’t it obvious? My G'MIC fun with afre thread almost exclusively deals with gradient_norm. So, any issues I have had with it are thorns in the flesh. Not necessarily improve but questing to find a better solution; my princess might be in another castle.

I was debating whether to put this in the processing category or gmic. I use gmic so gmic it was. But definitely applicable elsewhere like GIMP and Krita.

David_Tschumperle · January 28, 2019, 11:48am

interesting algos that could actually have their place in G’MIC!
Anyone interested to convert them as G’MIC scripts ?

afre · June 21, 2019, 7:34pm

Now that I have Octave, here is an actual comparison between gradient_norm (G’MIC) and phasecong3 (Kovesi). phasecong3 also outputs corners. Note that it contains many parameters, so I am sure the output could be optimized better.

canon3.bmp (converted to png using G’MIC)
canon3