Handwritten Digit Classification using SVD

Classifying 28x28 images by projecting them onto orthogonal bases

Drawing Pad

Draw a digit (0-9). Take your time, use multiple strokes if needed.

Singular Vectors k = 5

Higher 'k' uses more eigen-digits to reconstruct the image. Too high = copies noise. Too low = lacks detail.

Prediction

Draw a digit...

Residual Error || x - x̂ || (Lower is Better)

28x28 SVD Reconstructions (x̂)