A simple face detector for the Simpsons Characters. Used on seasons 25-28 it created a database of 9,878 images.
Each image:
- is focused on a single face
- has size 200x200
- is paired with a simplified version of itself, which only presents the skin and eyes of the detected face.
get_faces.ipynb gets the link of a directory where the video files are stored and extracts 150 images from each episode. Subsequently, it calls find_face() from face_detector.py which applies a variety of filters on each image in order to detect a face.
The process is based mainly on identifying clusters of white pixels that look like eyes, because the faces of the characters do not follow a consistent enough pattern.