trying to get better ede detection

While earlier we had tested
image -> sobel filter output -> grayscaling -> thresholding (-> median filter output) -> region partitioning,
on reordering the transforms applied as:
image -> median filter output -> sobel filter output -> grayscaling -> thresholding -> region partitioning
we find that
1. the number of broken edges is reduced resulting in better partitioning.
2. However, detail present in the thresheld and partitioned outputs is significantly low as well.
3. The second method works well even for very low values of threshold (23 was used instead of 63).
Detail may be added in preprocessing or the result, using detail obtained by wavelet transform. This is yet to be tested... for now it is time to go visit our guide and ask him for suggestions for localized spatial clustering, things like kernel methods in k-means might actually be relevant and will be assessed.

Prateek