如何在RGB图像中的阴影像素中消除阴影的影响

Question

I have high resolution ground cover RGB images with varying levels of shadow present. How can I pre-process them to remove the shadow effects (enhance shadowed pixels). Shadows mainly arise from rocks, tripod, and plants. Below are some examples (screencaptures of the actual images).

image1 , image2 , image3

After a second look - it appears that it would be very beneficial if the tiny shadows (eg created from small leaves, branches, small rocks) were also enhanced - I assume the solution would pick these up as well.

I'm doing a model of ground cover segmentation with three classes. The model copes well when light shadow is present so I'm not worried about that. The main concern is moderate levels of shadow. I basically want to feed the hundreds of images through some code so that the final result is an RGB image with the shadowed pixels appearing to not be in shadow anymore.

Something that increases the brightness of the shadowed pixels to match the mean brightness of the non-shadowed pixels might be a good starting idea (but I'm sure better techniques exist).

For the very dark shadows I plan on creating training data for a fourth class so I can detect these and remove these pixels from analysis (Assuming that very low brightness will be the main feature used here).

Answer 1

Maybe something trivial as converting the color space to YUV and setting an average Y brightness would do. Or something like the Retinex algorithm to break off the illumination layer (including shadows). Otherwise there are also several local enhancement methods like CLAHE that may help your problem. These are my suggestions.