My research repository. It is about decomposing face images into lower space.
- Wrote NMF algorithm myself
- Wrote algorithm to get a general inverse matrix
- wrote eigen decomposition algorithm
- Better algorithms for determing K values
- Reading on: motion detection
Dec 19: PCA Prototype.
- I wrote my own version PCA.
- Uploaded related results
Dec 21: Prototype of NMF, PCA
- Picked up NMF
- Using 'Moore-Penrose pseudoinverse of matrix' Algorithm for projecting results into NMF subspace, got great results.
- I found NMF doesn't explicitly restrict the dimensions of the subspace. I.e. For PCA, the matrix L is 400x400 thus at the maximum I can only get 400 eigenvectors. For NMF, due to its iterative, I felt I can go beyond that limit, it is not suggested. The matlab version limited this.
- NMF
- PCA
- Result images: I results I got so far.
- Research notes: I kept all important notes in here