To morph more the 2 faces, the main idea is totally the same.
Here I use a pre-labeled dataset containing faces of 25 women. Each face has 46 pre-labeled reference points, and I add 4 more points to each image at the 4 corners, so we have 50 points for each face in total.
Click the picture to access the dataset I use here!
Samples of a group of 25 women faces
(We will use aligned gray version, which has already been labeled, for analysis )
To produce the average shape, I used Delaunay Triangulation based on average coordinates of the 25 sets of reference points and put the triangles on to the average coorinates (warping fraction = 1/25) on the middle image.
Then same as before, I produce an average face by warping every face into the middle images and put average of their pixel colors (dissolving fraction = 1/25) back to the pixel on the middle image.
Mean shape
warpped to mean shape
original
To produce average face, first I wrap all the 25 faces into the mean shape respectively.
example 1
example 2
original
warpped to mean shape
Then same as before, I dissolve/overlap (dissolving fraction = 1/25) all the 25 warped results together to get the mean face. As we can see, compared to the mean face of the whole sample population of 100 people with mixed gender, the mean face in my example has more obvious female features.
Mean face of 25 women
Mean face of 100 mixed gender poeple
(given by dataset)
Without dissolving colors, we can also change 1 face into a second face's shape by only warping the face. Here I find the shapes (corresponding reference points + same DT) of my face and the mean face, and I can change my face to mean shape or change mean face to my shape by mapping the original pixel to the corresponding coordinate in corresponding triangle in second image.
Me
Mean
Me to mean shape
Mean to my shape
Me to mean shape
created with
Website Builder Software .