Deep-Fake

Full video: Deep Fakes

Here there are two files, one which takes 2 images a input and applies first image on top of second and second file where it takes 1 image and applies on the face of person captured from the webcam.

To achive this, we use shape_predictor_81.dat file

Using that we plot the facial landmarks on the image from webcam.

After this we construct delaunary triangles on that face.

We save all the points forming the face.

These same triangles are constructed on the image we provided and dividing the image into these triangular parts.

Now these triangular parts are reshaped to fit on same triangle from first image.

Now we reconstruct second image's face on first iimage.

Using openCV functions we make necessary color change to look more natural and realistic.

Future Scope:

• We can take a video of someone who is moving this face, then divide it similarly into delaunary triangles and story every image into folders inside another folder.
• Here we can store some lenghts. (distances of some important points to identift face orientation)
• Once its done, we can take imput from webcam and calculate the distances and pick the most matched image from our folder.
• Using that matched image we can construct much better image without weird distortions.
• Giving us much better results when face is rotated.