This searches for faces within an image and replaces them with the first face found from an image of your choosing. Some crummy bash scripts have been written to help convert video to and from images.
sudo apt-get install python-opencv
git clone git@github.com:matthewfranglen/face-replace-python.git face-replace
cd face-replace
python ferris.py REPLACEMENT_IMAGE REPLACEMENT_IMAGE INPUT_FOLDER/ OUTPUT_FOLDER/
This will convert all images in INPUT_FOLDER. It will replace any faces with REPLACEMENT_IMAGE and save the updated image to OUTPUT_FOLDER.
The replacement face can be detected from one image and then taken from another, allowing you to alter the replacement face (such as masking it).
python ferris.py DETECT_IMAGE REPLACE_IMAGE INPUT_FOLDER/ OUTPUT_FOLDER/
Here the DETECT_IMAGE is used to find the face, and then the same area is copied from the REPLACE_IMAGE. It is recommended that these be the same size. The size and position of the detected face is not scaled or moved to match the dimensions of the replacement image.
You can use the scripts in the bin folder to handle videos. This requires ffmpeg. The output video is saved in the mpeg4 format.
bin/strip INPUT_VIDEO INPUT_FOLDER AUDIO_FILE
python ferris.py DETECT_IMAGE REPLACE_IMAGE INPUT_FOLDER/ OUTPUT_FOLDER/
bin/combine OUTPUT_FOLDER AUDIO_FILE OUTPUT_VIDEO
This extracts the video at 25 fps so conversion can take some time. A two minute video is 3,000 frames.