"Layer-based" salient object detection

Testing a simple but fast masking technique for isolating the subject in captures made with scAnt

Approach

1. Convert the source image into HSV and LAB colour spaces
2. Detect which areas are in focus using Sobel filters on the L channel and S channel separately for rough and precise (resp.) detection
3. Use the H, S, L, A and B channels to isolate "layers" with dark areas, coloured areas, high light areas, etc.
4. Clip the upper or lower bounds in each layer to get rid of unwanted noise/ghosting effects
5. Merge it all back and binarise