We'll apply PCA to reduce the size of black & white images before applying classification algorithms:
Face Recognition
data: images of famous people goal: compress images and use them as samples for a classification task
We will work on a dataset of multiple B&W images in order to find common patterns between all images in order to reduce the number of "principal features" that describe them!
More precisely, we will try to express each image of our dataset as a linear combination of principal components using PCA.
In order to compress our images, we will then zero-out the smallest principal components and keep only the most important ones in the equation. Each "reduced linear combination" will represent an image that has been compressed. (Because we only removed the least important components,) Our lower-dimensional projection of the dataset will preserve the maximal data variance between images, so we should still be able to recognize which person is in each image.
