Project Overview

Building hybrid images and learning about filters

Part 1. Defining Correspondences

The first correspondence I wanted to build with a picture of Hugh jackman and Wolverine. They have never been seen in a room together and I think they look quite similar.

I used the project point correspondences tool given in the spec and used the Delauney mentioned in the lecture.

For the dissolve fraction and the morph fraction I experimented with several time equations, i ended up settling for the following quadratic:

time = t/frames

frac = s*time**2 + (1 - s)*time

With dissolve and warp s = 0.6, -0.2 respectively. (chosen experimentally)

Image 1
traingle mesh
Image 2
corresponding traingle mesh

Part 2. Computing the "Mid-way Face""

To compute the midway face I created a function that finds the weighted average of the shape (based on the warp fraction), then applies the same triangulation to the new shape. Finally for each triangle T_m in the mid-way face I solved for the transformation matrix A s.t. A*T_m = T_i. Using this approach to solve for the midway face requires interpolation; I tried a handful of interpolation functions, including bilinear as was suggested, but ended up using nearest neighbor because it’s the fastest and has good quality. Finally to compute the mid-way face we simply cross dissolve the colors between the both mid-way transformation:

Image 1
hugh
Image 2
morphed into the midway shape
Image 1
wolverine
Image 2
morphed into the midway shape
Image 1
wolverine
Image 2
midway shape + cross fade
Image 2
morphed into the midway shape

Part 3. The Morph Sequence

Image 1
hugh - wolverine
Image 2
me - forest
Image 2
george - me

Part 4. The "Mean face" of a population

average face shape of the whole population:

Image 1
average shape
Image 2
average face

Examples of morphing each of the faces in the dataset into the average shape:

Image 1
1a
Image 1
2a
Image 1
3a

average faces

Image 1
population
Image 1
morphed population

my face + average

Image 1
my face warped onto the average geometry
Image 1
The average geometry warped my face

Charicatures

I produced caricatures of my face by extrapolating from the population mean using: my_face_correspondance_vector + alpha(average_face_shape - my_face_correspondance_vector)

Image 1
alpha = -1
Image 1
alpha =1.5
Image 1
alpha = 0.25

Bells, whistles and all things

ells, whistles PCA Basis

With the same dataset I computed the PCA Here are the top 5 eigenfaces. Very spooky

Image 1
1
Image 1
2
Image 1
3
Image 1
4
Image 1
5

The top 5 faces capture a lot of information so they look like reasonable faces unlike the smaller eigenfaces which degenerate. I also tried my hand at projecting faces onto the eigenfaces (to get the weight) and then recovering them. The following are examples with the top 20 eigenfaces. Even with few eigenfaces a lot of essential information is recovered.

Image 1
1a
Image 1
2b
Image 1
181b

Bells, whistles Gender Bend

Change gender/ethnicity of my friend's face. Using this image https://pmsol3.wordpress.com/2009/10/10/world-of-facial-averages-east-southeast-asia-pacific-islander/

First just the shape:

Image 1
1
Image 1
2
Image 1
3

Just the apperance:

Image 1
1
Image 1
2
Image 1
3

Both

Image 1
1
Image 1
2
Image 1
3

Moprphing with Janhvi Kapoor also worked quite well:

Image 1
1
Image 1
2
Image 1
3

Bells, whistlesRoomate morph music video:

https://youtube.com/shorts/FU-OLUDE05s?si=19tKs5OtATjvTVrH

Information

This website contains transitions not captured by the pdf, spesificaly, the title image changes into a high gamma verison and then into the black and white threshold filter version.