Compressing and enhancing hand-written notes

image-processingk-means-clusteringcolor-quantizationpythonpdf

Abstraction: k-means color quantization pipeline to clean and compress handwritten note scans

Key points:

noteshrink.py reduces 7.2 MB scanned PNG pages to ~121 KB while improving visual clarity by removing bleed-through and noise
Background color identified by finding the mode of a 5% pixel sample after reducing bit depth to 6 bits/channel to create robust bins
Foreground isolation uses HSV colorspace thresholding (value diff > 0.3 or saturation diff > 0.2 from background) rather than RGB Euclidean distance, correctly handling bleed-through vs. ink
Foreground color palette chosen by k-means clustering (k=7 clusters for 8-color output) in RGB space; each pixel mapped to its nearest cluster center
Final output is an indexed-color PNG, further reduced by running optipng/pngcrush/pngquant; pages assembled into PDF via ImageMagick
Four example PDFs averaged ~130 KB/page output; tool is on GitHub as mzucker/noteshrink

Connections: Color Quantization · K Means Clustering · Image Processing