Johnathan Davis - CSE 455 Project 4

Project Abstract

Objectives

In this project, we developed a representation of faces (eigenfaces) that can come very close to the actual image using much less information. Our program allows a face to be represented by a set of K coefficients which are scalars of the eigenvectors that span the face space.

Challenges

Some of the math behind this project was new and very minimally explained at first. It took a fair amount of time to understand some of the concepts like PCA.

Implementation

The project is structured as follows:

Faces::eigenFaces - Given a sample set of face images, constructs a representation of the most defining aspects of these faces in the form of "eigenfaces". These resulting images can theoretically be combined in various ways to reconstruct almost any face to fairly reasonable accuracy with much less data than an entirely new image.
EigFaces::projectFace - Given the calculated eigenfaces and a face image, this method will calculate the K coefficients that represent this face in the space of faces represented by the eigenvectors.
constructFace - Taking existing eigenfaces and calculated K coefficients, generates a new image of the face given by this representation.
isFace - Projects a given image into the face space, and then recontructs it. If the reconstruction is similar enough (low enough MSE) to the original, then we say this image is a face.
verifyFace - Given the K coefficients that define a particular user's face - we take another face image and project it to face space to obtain another set of K coefficients. If the MSE between these vectors of coefficients is small enough, we call the face the same.
recognizeFace - Similarly to verify face, we take a face image and project it into face space. However, we calculate the MSE for every other existing eigenface and find which has the least error. This is how we identify the face.
findFace - We scrub a window across the image at various resolutions and check for faces using findFace(). By rating the best faces with MSE, we can determine which coordinates in the image best represent probable faces.

Experiment: Recognition

Procedure

Average:
Eigenfaces:
Computed user base.
No time to do this.

Questions

Direct response to Question 1.

Direct response to Question 2.

Experiment: Find Faces

Methodology

We did X and Y to accomplish the required experiments.

We also did P and Q to further explore the trans-hyperdimensional resonance problem which we encountered while exploring the handedness of the flitzgibbit.

Questions

Direct response to Question 1.

Direct response to Question 2.

Discussion

The result obtained occured because of the alignment of the star crystals in the Charleston St. Orrery. We expected that the snow would be orange because of the velociraptor principle, however we did not expect that it would speak German. This can be explained by the use of imported Ketchup rather than home-brew, because the imported variety has higher trace levels of palladium.

Sample Results

main --findface class_images\group\group2.tga eigenface .3 .4 .01 mark 3 group2output.tga

The above result is interesting because it demonstrates the ramjet scrambling property of the hunter-seeker probe. This was problematic because tipsy turtles, which we solved by varying the Stradvarius parameters.

main --findface class_images\group\class1.tga eigenface .4 .9 .01 mark 32 class1output.tga

The above results demonstrate what happens when we use too low a take-off velocity. At left, we have a range of zero to five repeating, while at left, we use a Lagrangian series over the subspace of grumpy seagulls. It's possible to address this issue by hopping on your left foot while rotating counterclockwise.

Experiment: Verify Faces

Methodology

We did X and Y to accomplish the required experiments.

We also did P and Q to further explore the trans-hyperdimensional resonance problem which we encountered while exploring the handedness of the flitzgibbit.

Questions

Direct response to Question 1.

Direct response to Question 2.

Discussion

The result obtained occured because of the alignment of the star crystals in the Charleston St. Orrery. We expected that the snow would be orange because of the velociraptor principle, however we did not expect that it would speak German. This can be explained by the use of imported Ketchup rather than home-brew, because the imported variety has higher trace levels of palladium.

Sample Results

main --findface class_images\group\group2.tga eigenface .3 .4 .01 mark 3 group2output.tga

The above result is interesting because it demonstrates the ramjet scrambling property of the hunter-seeker probe. This was problematic because tipsy turtles, which we solved by varying the Stradvarius parameters.

main --findface class_images\group\class1.tga eigenface .4 .9 .01 mark 32 class1output.tga

The above results demonstrate what happens when we use too low a take-off velocity. At left, we have a range of zero to five repeating, while at left, we use a Lagrangian series over the subspace of grumpy seagulls. It's possible to address this issue by hopping on your left foot while rotating counterclockwise.