Face Obfuscator is a chrome browser extension that uses face detection and recognition algorithms to dynamically and locally block pre-trained faces from images loaded in chrome. The extension is powered by the face-api javascript library. Images are initially loaded with a gaussian blur, then obfuscated by dynamically adding a red rectanglur box over the recognized faces.
-
Open the Extension Management page by navigating to chrome://extensions.
-
The Extension Management page can also be opened by clicking on the Chrome menu, hovering over More Tools then selecting Extensions.
-
Enable Developer Mode by clicking the toggle switch next to Developer mode.
-
Click the LOAD UNPACKED button and select the extension directory.
Note: The extension is slow on iniital navigation to website.
Scripts and styles (content.js, face-api.min.js, and styles.css), are injected and loaded on every webpage, which drive the manipulation, detection, and obfuscation of images.
Face descriptors were generated from 300 images under labeled_images and saved in a descriptors.json file. The images were collected from a subset of a Kaggle dataset. The descriptors computed from these images are loaded on every site navigation. Improvements to this injection to speed the dynamic manipualtion should be explored. There are also some errors on continually applying the face detection and recoginiton algorithims on dynamically loaded images. This continual application is driven by the javascript mutation observer.
We also modify the 'access-control-allow-origin' header from requested images in order to bypass CORS protocol in background.js. This allows us to read and modify images downloaded from other origins in the HTML5 canvas.
The extension was developed as an academic project and we intend for this project to be more as an educational resource, learn by open sourcing. We are open for feedback and contribution.