[scikit-image] GLCM for non-rectangular image

[Juan Nunez-Iglesias]

Hi Kenlo,

Thanks for your email.

`np.nan` is only defined for floating point images, but GLCM only works for non-negative integer images, since the intensity values need to serve as indices to a matrix.

Therefore, in order to ignore NULL values, you just need to create an array with an arbitrary value as NULL. This is actually not too hard. If you start with a uint8 image, make a uint16 image with a mask where values to be ignored are set to 256.

Then, use greycomatrix with levels=257.

Finally, from the documentation:

>        The grey-level co-occurrence histogram. The value
>        `P[i,j,d,theta]` is the number of times that grey-level `j`
>        occurs at a distance `d` and at an angle `theta` from
>        grey-level `i`.

Thus you can get your odd-shaped GLCM, ignoring values outside your mask, with:

out = greycomatrix(image, […], levels=257)
result = out[:256, :256]

I hope this helps!

If you have suggestions for making this process easier, please let us know!

Juan.

On 7 Sep 2017, 11:23 AM +1000, NASAHARA Kenlo <24dakenlo at gmail.com>, wrote:
> Dear scikit-image community:
>
> I am trying GLCM functions in scikit-image.
> http://scikit-image.org/docs/dev/auto_examples/features_detection/plot_glcm.html
>
> I wonder whether they can treat images with irregular
> shapes, such as a circle, an oval, or some more complex
> forms (not only rectagle).
>
> In the "greycomatrix", function:
>
> def greycomatrix(image, distances, angles, levels=None, symmetric=False, normed=False):
>
> image : array_like
> Integer typed input image. Only positive valued images are supported.
> If type is other than uint8, the argument `levels` needs to be set.
>
> the "image" seems to be required to be a rectangle.
>
> If NAN or NULL values are allowed in the "image", and the function skips
> treating those "no-data" pixels assigned with NAN or NULL, then we
> may be able to treat any shapes of image, by filling the pixels outside the
> boudary with NAN or NULL.
>
> However, in the source code of "_glcm_loop" function
> (https://searchcode.com/codesearch/view/83509141/), all the pixels
> within the rectangular image are scanned, without permitting "no-data".
>
> My questions:
> 1. Do you know how to treat GLCM in non-rectangular image with
> scikit-image?
> 2. If not, do you know any other ways to do that instead of scikit-image?
> 3. Do you think it is a reasonable idea to change GLCM functions in
> scikit-image so that they can treat no-data (allowing NAN or NULL)?
>
> --
> Kenlo Nishida Nasahara
> Faculty of Life and Environmental Sciences
> University of Tsukuba, Japan 305-8572
> 24dakenlo at gmail.com
> _______________________________________________
> scikit-image mailing list
> scikit-image at python.org
> https://mail.python.org/mailman/listinfo/scikit-image
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-image/attachments/20170907/3e8068b2/attachment.html>