Instructor Onboarding for Data Carpentry: Image Processing with Python (beta)

Robert Turner, “Image Processing with Python” maintainers

August, 2022

Thank you!

Thanks to CarpentryCon 2022 for having this session, thank you for coming.

Presenters

• Robert “Bob” Turner
• Toby Hodges
• David Palmquist

Acknowledgements

• Special thanks to Prof. Mark Meysenburg, Prof. Tessa Durham Brooks, Dr. Raychelle Burks, and everyone involved in the initial design and testing of the curriculum within the DIVAS Project.
• Dominik Kutra and Dr. Constantin Pape made an essential contribution by converting the lesson to use scikit-learn.
• Dominik Kutra and Dr. Gregor Mönke taught the alpha pilot at EMBL, with help from Dr. Christian Tischer and support from the EMBL Bio-IT Project.

Acknowledgements (continued)

• Trisha Adamus and Prof. Mark Meysenburg taught the alpha pilot at the University of Arizona, with help from Zuzana Adams, Amirhossein Azami, Ryan Carlson, Courtney Comrie, Gabriela De La Cruz Sanchez, Chris Klimowski, Shuailong Li, Chuan Luo, Artin Majdi, Maliaca Oxnam, Travis Struck. That workshop was hosted by the BIO5 Initiative, in partnership with CyVerse, the D7 Data Science Institute and the UArizona Libraries

Acknowledgements (continued)

• Dr. Kimberly Meechan, David Palmquist, Prof. Ulf Schiller, and Dr. Robert Turner have taken over as Maintainers and core contributors for the lesson, and continue to drive its development.
• Many community members have opened issues and pull requests to improve the lesson.
• The lesson development process was supported by Dr. Erin Becker and Dr. Toby Hodges from The Carpentries Curriculum Team.

What is a “beta” lesson?

Pre-alpha–>Alpha–>Beta–>Stable

• Exercise caution
• Feedback welcome
• Second beta pilot needed (please get in touch)

Why teach this lesson?

• Images are everywhere.
• Images data is different to data frame data.

What’s in the lesson?

• Introduction to images in research.
• How images are represented by computers.
• Software tools for working with images.
• Manipulating images using software.
• Extracting data / statistics from images.

How do we think about image data?

Key concepts

• Pixels
• Arrays
• Coordinates
• Channels
• Kernels
• Binary masks

What’s not in the lesson?

• AI / Deep Learning

Lesson is a more “traditional” approach: easier to explain results, less data-intensive.

What do I need to know first?

• Bash shell skills
  • Navigating directories using pwd, ls, cd <subdirectory>, and cd .., Run a Python script from the command line.
• Python skills
  • Variables and types, lists, logic (if, else, etc.), basic file input / output

Setup

• Data
  • Download from FigShare
• Software
  • Anaconda (base environment includes all required packages) and Jupyter Notebooks

Ep 1. Introduction

• What research questions can we answer with image processing?
  • e.g. imaging a Black Hole, estimating the population of Emperor Penguins, the global-scale analysis of marine plankton diversity, segmentation of liver and vessels from CT images
• Morphometrics

Optional Breakout 1

• What research area are you most comfortable in?
• What research areas do you expect your learners to come from?
• Are there any particular challenges in working with image data in these areas?

Ep 2. Image Basics

• Representation of images in computers.
• Images, arrays and pixels.
• How RGB is used to make colour images.
• File formats and compression.

If you are a computer, images are arrays

Image with pixel values overlaid

Image Representation

RGB Image

Red channel

Green channel

Blue channel

Ep 3. Working with skimage

• skimage (Scikit Image)
  • Input + Output (images are not stored in numpy friendly formats)
  • Do weird things with arrays (e.g. skimage.transform.resize)
• numpy
  • Conventional array stuff (e.g. sub-setting, find all values > x)

Ep 4. Drawing and Bitwise Operations

• Creating images from nothing
• Masking (setting pixels to be ignored)

Ep 5. Creating Histograms

• The distribution of intensity of colour in an image can tell us things.

Plant Seedling

Histogram

Ep. 6 Blurring Images

• Why blur?
• What’s the relationship between blurring, filtering, convolution and kernels?

Effects of blurring

Not Blurred

Blurred

Ep. 7 Thresholding

• What is thresholding?
• Why would we want to do it?
• What methods are available?

“Simple” Thresholding

Blurred Grayscale

Threshold Applied

Ep 8. Connected Components

• Separating objects and getting information about them.

Labelled Shapes

Getting statistics

Areas Histogram

False Positive Objects

Morphometrics

• Properties of the shape of an object.
• skimage regionprops
• Basic e.g. area, perimeter, center
• More complex e.g. eccentricity, bounding box

Ep 9. Capstone Challenge

• Brings together blurring, thresholding and connected component analysis.
• Example used is counting colonies of bacteria.

Optional Breakout 2

• What questions would you expect to be asked by learners?
• How would you answer them?

Questions / discussion

Thanks again!