Facial Recognition Using KL Decomposition
This is a group project I worked on in Mathematical Modeling Case Studies. We were given two weeks to complete this project.
In this project the strengths and weaknesses of image-based facial recognition software by means of Karhunen-Loeve decomposition (KLD) were tested. KLD is a type of Singular Value Decomposition (SVD) and has been been used for facial recognition software since 1971. The images were all normalized by FaceTool 2.0. Eigenfaces were created from a training set and used to reconstruct images in a test set. The Euclidean distance of the resulting coefficients was then used to compare the similarity between reconstructed images. With a threshold of above 58% similarity, there were no unexplained false positives.