tongsong91 / average_face Goto Github PK

This is a bunch of simple scripts that scrape the Oxford Experimental Psychology website for people's personal photos. These can be uploaded by each individual member of staff, and by grad students who are working towards their PhD. There is no standard format required (I think there might actually be some rules, but most people seem to ignore them), which means that the photos need to be automatically processed to find the faces. For this, I used the Haar cascade algorithm from @shantnu's FaceDetect project. After all the faces are detected, they are averaged, and the resulting image is saved. (And it might haunt your dreams for years to come.)

Note that not all the faces have the same orientation, so the averaging is imperfect at best. One solution could be to detect the eyes, and realign each image to a horizontal eye-line. Have fun implementing that, and give me a shout (and a pull request, please) once you're done.

1. Run extract_faces.py. This will scrape the 'team' part of the EP website. The html source of this page contains references to all EP members. This script finds the links to each member's image, and then downloads those. The downloaded images will be in a folder called scraped_faces.

2. Run detect_faces.py. This will go through all the images in the scraped_faces folder, and it will attempt to detect any faces in those images. Note that it can detect multiple faces per image, and that it can make mistakes every now and again. The detected faces are cropped, and the resulting images are stored in a folder called detected_faces.

3. Manually check the faces. This step is a manual one, because humans are waaaaaaay better at detecting faces than computers are. Create a folder called filtered_faces, and copy all the images from detected_faces that actually contain a face.

4. Run average_face.py. This will load all images in the filtered_faces folder, and it will then calculate the average face. This will be stored as EP_average_face.jpg.