Fun with Filters and Frequencies

Methodology

First, I cropped and resized an image of myself and my friend Mckenzie to ensure both had the same dimensions, with our faces matching in size and aspect ratio. Using the provided correspondance tool, I selected 58 correspondence points for each image including the four corners to make sure that all triangles were created. Then I generated a Delaunay triangulation based on the average of the two sets of correspondence points.

Part 2: Computing the Mid-Way Face

Blurred Finite Difference

To compute the mid-way face between me and Mckenzie, I followed these steps:

1. Calculated the average shape by taking the midpoint of the corresponding points

2. Computed the inverse of the affine transformation matrix for each triangle, mapping the original triangles in both images to the corresponding triangles in the average shape.

3. Applied inverse warping using a polygon mask for all the pixels.

Then I cross-dissolved the warped images by averaging them together.

My original image

Midway

Mckenzie's original image

Part 3: The Morph Sequence

Using Part 2 and varying the weights between 0 and 1, I created a morph sequence consisting of 56 frames.

Part 4: The Mean Face of a population

To compute the mean face of a population, I took the FEI database for Brazilian faces and used the images of a straight face to try and find the average face. To find the average face, I followed these steps:

1. Calculated the average shape by taking the mean of all corresponding points.

2. Warped each face to fit this average shape.

3. Compute the mean face by averaging the colors of all the warped faces.

Below are examples of faces from the dataset that have been warped to the average shape

1: straight face morph

5: straight face morph

6: straight face morph

1: smiling face morph

5: smiling face morph

6: smiling face morph

Average straight face

Average smiling face

Here is my face warped into the average geometry, and the average face warped into my face’s geometry.

My face warped

Average face warped

Part 5: Caricatures & Extrapolating From the Mean

Using the FEI database population means, we can create caricatures by extrapolating facial features. I used p=0.5 to create the caricature meaning it is halfway between two sets of facial correspondences

Bells & Whistles

For my initial enhancements, I transformed my features with that of an average Chinese male face that I sourced online. The process began with rescaling and resizing both images to ensure compatibility, followed by selecting corresponding points between them. I then explored three distinct morphing techniques:

1. Warp: Matching the structure of the two faces
2. Cross-Dissolve: Matching the appearance of the two faces
3. Warp and Cross-Dissolve: Matching the structure and appearance of the two faces

Average chinese male

Warp

Cross-Dissolve

Warp and Cross-Dissolve

Introduction