Comment by flobosg

> How were the images processed?

The methodology is described further down in the article. The author extracted features from the screenshots with OpenCV and then reduced the dimensionality of the feature set using t-SNE.