The rationale for the approach this page takes is elaborated on Edge Detection Scripts.

The software looks at detecting edges from an image.

Different images and parts of images are different. Some have very stong contrast in brightnesses and others less so, some are more delineable from their contrasting hues, some from a contrast in colour saturation. The classic edge detection software such as the formulas devised by J.F Canny in 1986 work by detecting levels of brightness of adjoining pixels. This is undoubtedly a very strong factor in identifying an edge, but it isn’t the only one, we see an edge if there is contrast in hue, too, even two hues of the same brightness, provided that is that the brightness is quite high, though not too high. We won’t detect an edge if the adjoining hues are very dark, nor if they are very bright.

And while we can detect an edge if two adjoining pixels are of different brightness, the edge detection wil be more apparent with a greater difference in brightness.

Unlike the classic edge detection software my software does not work on gradients. Instead it looks at a specified number of pixels verically above, horizontally to the left, and diagonally north-west and takes an average of the difference in that number of pixels to calculate relative brightness, as regards hue it similarly works on average difference, using perceptual hue name, rather than numeric value on the colour circle.

You’ll see that painting a result from the default parameters on this page produces a lot of lines. This is on purpose. It is not to make a pretty image, it’s to try and isolate by edges the various different shapes in the original photo.

The coding scripts used by this page are all client-side Javascript. It works reasonably fast, I am optimising all the time, could be faster though because of the volume of data being handled. Pushing the processing to the server does not seem all that helpful, though as with all my software I am constantly experimenting.