Feature Compression using PCA and SVD in Images from Scratch

• Performed lossy compression using PCA and using SVD from scratch on images with varying number of principal components.
• Generated Eigen Faces from Olivetti dataset using PCA with different number of eigen values.

Dataset

• Olivetti Face Images: 400 images of 40 distinct people, each 64x64 grey-scale images.

Output with Top-123 Eigen Values: 5% Error

Output with Top-5 Eigen Values

Error vs Number of PCA Components Considered

Contributions

Saurabh Chatterjee
MTech, Signal Processing and Machine Learning
Indian Institute of Technology (IIT) Kharagpur