gradient magnitude options

ofgulban / segmentator

3D MRI data exploration and segmentation tool.

https://github.com/ofgulban/segmentator/wiki

BSD 3-Clause "New" or "Revised" License

74 stars 13 forks source link

gradient magnitude options #47

Closed ofgulban closed 7 years ago

ofgulban commented 7 years ago

Gradient magnitude can be computed in several ways:

Central differences in the interior and first differences at the boundaries, aka numpy.gradient
with Sobel operator
with Prewitt operator

We are currently using numpy.gradient but I am planning to change the default to Sobel operator because it is less sensitive noise. Changes are currently implemented in devel branch.

Gradient magnitude comparisons: gramag_animated

ofgulban commented 7 years ago

After reading a bit more about this (reference 1, reference 2), it seems that Scharr kernel is more rotation invariant compared to Sobel (Kroon, 2009, see figure 3) therefore it should be an improvement over Sobel.

I have implemented the 3D Scharr operator with appropriate scaling and made it the default option in v1.3.0.

See the gradient magnitude comparisons with 2D histograms attached: animated (grayscale range is the same across images)