Image Classification with PCA

Use principal component analysis for image classification.

← Linear Algebra projects

Learn Mathematics Linear Algebra

Question

How can an image be compressed while keeping the features needed for classification?

PCA keeps the strongest patterns in the data and removes many less useful dimensions.
PCA keeps the strongest patterns in the data and removes many less useful dimensions.

Goals

  • Find principal components from a collection of images.
  • Project, reconstruct, and classify images in a smaller space.

Method

Replace thousands of pixel values with a few directions that explain the most variation.

  • Turn each image into a vector, center the data, and compute a covariance matrix or SVD.
  • Keep the leading eigenvectors and project every image onto them.
  • Train a simple classifier and measure how the number of components affects accuracy.