Currently 2 fixed borders are used in the HSV color space
But this does not deliver good results in every case. Bright colors with low green saturation apear much more green than dark colors with the same saturation
There must be multiple borders to make the mask --> Solution is to allow more hues and apply an Otsu thresholding after the initial HSV Inrange operation
c476d2506e2cd9e298317ca9f5ff420de955c4f7 improved leaf segmentation algorithm.
See description above for details. Basically it checks first for color for a basic segmentation and then applys otsu thresholding. Otsu thresholding should be applied in trivial cases between black and saturation >0.
In other cases between 2 groups of saturations.
HSL calculation is now fixed aswell. The image is read as BGR and the channels must be interpreted that way
0940f7e90ccfa44cef1201183f9a71536bebd583 added pixelcount as descriptor for better checking whether the results are correct.
The improved segmentation seems to also improve the leaf counting
a8e2d284409eee9e18a28eb614abb57c26254243 Only temperatures above zero degree celsius are considered for temperature calculation
a95c980cdc1c462e98de2dce0a2e97cd3d5e6559 Applying opening after otsu thresholding again to remove noise left after thresholding.
On some images this does have an effect.