This was a project completed in the CS251 course at Colby College. The main goal of the project was to use PCA as dimensionality reduction for images. However, I also used PCA as a method of simple facial recognition.
The dataset used was the LFWcrop Face Dataset (https://conradsanderson.id.au/lfwcrop/), which is a cropped version of the Labeled Faces in the Wild Dataset (http://vis-www.cs.umass.edu/lfw/). The data set "contains more than 13,000 images of faces collected from the web" and was originally "designed for studying the problem of unconstrained face recognition". Due to file size, this dataset was not uploaded, but can be found at the above link.
- Implement the Eigenface Algorithm on the LFWCrop
face dataset
- PCA uses a number of "principal components" to explain variance in data in order to reduce dimensionality
- Steps of PCA consist of standardizing input, computing the covariance matrix, using eigenvectors and eigenvalues to compute principal components, creating a feature vector from the top principal components, and projecting principal components back onto original dataspace. -For this dataset, the top 200 principal components were chosen
- Use PCA for Facial Recognition
- A image of an indivual from the dataset is chosen (a different image that is not in the dataset).
- project this image into the PCA space using the feature
- compute distance between the new image vector and each of the original image vectors
- return name of image with smallest distance to new image