Improve IT8 detection

[drtrigon]

Do you want to extract the color bar/scale on the bottom only or all the tiles? I was consider the color bar/scale only.

For the tiles I am wondering now whether they are standardized in size such that you could just extract a grid of points after doing some normalization and aligning.

[drtrigon]

For the bar/scale I am still mentaly sticking with segmentation methods may be also something like http://scikit-learn.org/stable/modules/feature_extraction.html#connectivity-graph-of-an-image

[AbdealiLoKo]

Not sure what tile means in your earlier comments - could you elaborate ?

Currently I have not found an IT8 Strip anywhere except the bottom and top of an image. Hence I'm focusing to detect only at those points.

Regarding segmentation: Possibly I'm not using segmentation correctly ... I agree that segmentation does seem like the correct way to solve it, but I'm getting unusual results as discussed. I should understand the underlying algo better and play with the parameters smartly.

[drtrigon]

tile: have a look at https://en.wikipedia.org/wiki/IT8#/media/File:IT8_color_target_by_EGM_Laboratories.jpg ... do you want to extract the gray bar at the bottom only or do you also want to extract the "tiles" A1 till L22 as well? That is what I would use a grid of points for.

About segmentation - I might guess low resolution could be a problem here (imagine a segment of 1px in width/height. May be it will help if you scale them up e.g. bilinearly.