Imager Update

Occlusion Table

I continued to tweak the code (and actually got it to work properly). After a brief break from the project, I realized how difficult it was to follow the code without extensive commenting… so, the updated code is extensively commented - at least from my perspective. I also moved the project folder to the SMAIG project and it now resides on GitHub.

Occlusion scores were derived by counting the number of pixels with a color code greater than .5 (0 = black and 1 = white in a gray-scale palette) - lower scores = greater occlusion. The images can all be regenerated using the SMAIG code and the spreadsheet in the folder.


Did you find this page helpful? Consider sharing it 🙌

Tilman Sheets
Tilman Sheets
Professor of Psychology

My research interests include automated item generation and assessment of cognitive abilities.