An experiment in detecting DoNotSnap badges in photos, to protect privacy.

This program allows you to detect and identify DoNotSnap badges via a sliding-window decision tree classifier (custom heuristics are used to reduce search space). The classifier is trained by matching samples against image templates using Affine-transform invariant SURF features.

You can find examples of using the classifier in classify.py and training a new classifier in train.py

A pre-trained classifier can be found in classifier.pkl Alternative versions of the same classifier are in classifier_alt_1.pkl and classifier_alt_2.pkl

Run python classify.py <path-image-to-be-tested> This will deserialize the classifier from classifier.pkl and run it on the image you supplied. A sample image could be found in sample.jpg

Run python train.py <output-file> <total-number-of-samples> This will read the sample filenames from positive.txt and negative.txt files. Templates filenames are specified in templates.txt. A sample template could be found in template.png The output is a <output-file>.pkl with serialized classifier.

• opencv
• numpy
• sklearn
• matplotlib
• PIL