In the first part of the project we are going to morph two face images. Firstly, corresponding points on two face images are manually choosen and the average shape which is basically the middle of corresponding points is calculated. Then we perform the trangulation on this average points and this trangulation applies to both original images. Affine transformations can be found for corresponding trangles and in this way, the shape of original images are wrapped into the average shape. By interpolating the value of pixels, we are able to cross disslove between image A and B.
The second part of the project aims to calculate the 'mean face' of a population using available dataset of annotated faces. With annotated data, an average face which is the average of corresponding points is first calculated and the trangulation is performed. All individual images are wrap into this average one and the colors are also averaged to get the final 'mean' face. The evolution from individul face to the average and from average to the individual will be shown. Furthermore, a caricature of individual is achieved by changing from the 'neutral face' to the 'smile face' in terms of shape, appearance and both.
| Image A: | Image B: |
|---|---|
|
|
| Mid-way face: | |
|
|
| Animation: | |
|
In this part, IMM Face Database from Danes used. 37 'neutral' faces and 37 'smile' faces are used to calculate a 'neutral' average and a 'smile' average.
| Average of 'Neutral' faces: | Average of 'Smile' faces: |
|---|---|
 |
 |
| Examples of morphing: | |
| Face1 to the averaged 'Neutral' face: | The averaged 'Neutral' face to Face1: |
 |
 |
| Face2 to the averaged 'Neutral' face: | The averaged 'Neutral' face to Face2: |
 |
 |
| Extrapolating the average 'neutral' face to Face1: | |
|---|---|
| Average 'Neutral' face: | Face1: |
|
 |
| Extrapolated face: | |
 |
In this part, an idividual 'neutral' face image is changed to a 'smile' one in 3 ways. First, it is morphed to the average shape of 'smile' faces. Second, the difference between average 'smile' face and average 'neutral' face is added to it. Third is a combination of the previous two.
| Original 'neutral' face: | Wrapped to the average 'smile' face: |
|---|---|
|
 |
| Difference between average 'smile' face and average 'neutral' face: | Added with the Difference: |
 |
 |
| Both: | |
 |
It can be seen that by wrapping the original 'neutral' face to the averaged 'smile' one, the person smiles a little bit more. However, the effect of just adding the appearance doesn't work very well due to the dissimilarity between the original face and the averaged 'smile' face. The 'smiling teeth' in the differece is obvious which is a clear sign of 'smiling'. By combining both the shape and appearance effect, the origianl 'neutral' face becomes a 'smile' face.