Show and tell: how are you using scikit-image?

[emmanuelle]

Hello skimagers! This issue is to ask you to share how you are using scikit-image. We would like you to explain your use cases and image processing workflows.

Why are we asking this, and how will the information be used? Mostly to know if our documentation is relevant to your applications, and how it could be improved.

Here is a possible template of information to share, but feel free to organize the information differently, to answer only some of the questions (we understand that there might be confidentiality issues and information you cannot disclose), or to provide other information.

• For which application are you using scikit-image? (Example: computing shape properties of blood cells in microscopy images, for the detection of pathology XX).
• Which organization are you working with? Who uses scikit-image within the organization?
• How has scikit-image helped us for your projects?
• Can you share links to code examples (preferably in online repositories such as Github, Gitlab, Bitbucket etc.), typical images from your applications, tutorials or research papers showing how you are using scikit-image?
• If you cannot share code, which functions of scikit-image do you use most often?
• In your opinion, what is missing in scikit-image features and documentation ?
• If we write examples related to your application, can we reach out for a review of the example (to know whether it is relevant)?

How to answer: just fill the next "Leave a comment" field. You will be able to update the field if you wish (click on the three dots in the bar, then Edit), so don't worry too much about making it perfect the first time.

Thank you so much for helping us to improve scikit-image!

[emmanuelle]

So let me start with the work done by @BBarbara-fr:

• Application: measuring shape properties of interconnected and disconnected objects in SEM images of phase-separated silicate glasses (materials science). The context is to control the morphology of phases for micro and nanopatterning.
• Organization: SVI lab, joint CNRS and Saint-Gobain lab in Paris. PhD students, post-docs and researchers are using scikit-image.
• Links to examples : https://github.com/svi-lab/segmentation/blob/master/interconnected_porosity_2D.ipynb. Related publications : https://www.sciencedirect.com/science/article/abs/pii/S1359645415002219
• Most used functions, pipeline : correcting uneven exposure with exposure.equalize_adapthist, denoising (median, denoise_tv_bregman), filters.threshold_otsu, mathematical morphology for post-processing, measure.label and measure.regionprops.
• What is missing: more examples using images from materials science. Morphological granulometry function.
• Available for reviewing examples.

[sciunto]

I gather several applications as they are using similar techniques.

• application: measuring morphological or positional properties of drops, fibers, bubbles from direct imaging.
• Organization: https://www.equipes.lps.u-psud.fr/mmoi/
• https://arxiv.org/abs/1610.02555 https://arxiv.org/abs/1605.05085
• function used: Hough transforms, ransac, thresholding (segmentation in general), labeling

[alexdesiqueira]

Hey everyone,
I have used skimage thoroughly since 2014, with several applications.
One of them is finding fission tracks in mineral images. We wrote an alternative watershed transform for that, WUSEM, published recently.

• [paper on JEMT][arXiv][code]

[carterbox]

Hi all,

I'm a casual skimage user and a maintainer of a python package for tomographic imaging. My applications are mostly processing of x-ray images from the Advanced Photon Source.

• We use skimage in our library for resizing 2D images and currently for tranformations and image warping.

• I have also used skimage.exposure for adjusting the dynamic range of x-ray images

• and also for deblurring images using skimage.restoration.

• I think I also used the phase unwrapping routine once.

[joefutrelle]

skimage is a key part of our workflow for segmenting microscopic images of plankton for the purpose of estimating properties such as biovolume. The workflow was originally written in MATLAB, and skimage's morphology operations allowed me to duplicate the output of the MATLAB code pixel-for-pixel.

In the process I had to implement a morphological thinning operation which I contributed and is now the skimage.morphology.thin function. Contributing to skimage was a very rewarding process and I appreciate how welcoming and helpful the team was as I prepared and documented my contribution.

We have over 1.2bn images of plankton in our database and we plan to run this code on all of them!

[SuborbitalPigeon]

We've been making use of scikit-image for processing ultrasound C-scan images with a view to detecting defects in composite aerospace components. We've written a light wrapper around the various processing functions to make stringing them together into pipelines easier.

[grlee77]

Came across this old issue today. It's great to hear how people are making use of scikit-image.

I just pinned this to the issues page to increase visibility for others how may also be interested in providing feedback on use cases.

[Anchaliya75]

@emmanuelle How it is different from OpenCV, I am new to scikit Image,Is it an alternative to OpenCV ?

[jni]

@Anchaliya75 the main differences between openCV and scikit-image are:

• scikit-image is written in Python, with some small parts written in Cython or Pythran for speed. OpenCV is a C/C++ library with Python wrappers. The three main consequences of this are (1) the API of scikit-image (the way functions are named, the types of the inputs and outputs, etc) is more "Pythonic", that is, closer to what you would expect in Python. (2) if you know Python, it is easier to hack, modify, and improve scikit-image methods yourself compared to OpenCV. But (3) typically, if a function is available in both libraries, the OpenCV version is faster.
• scikit-image targets scientific usage, and this means in general it has better support for 3D and nD images, while OpenCV has more focus on natural images (e.g. face detection in videos), so many parts work with only 2D images (but this may have changed recently, I haven't followed it closely).