Zero-Upload Perspective Document Scanner & B&W OCR Contrast Studio (2026)

When you photograph a paper page under indoor lighting, one corner of the white paper in shadow is often darker in raw RGB luminance than the black ink in the brightly lit corner. A single global threshold turns the shadowed half completely black. Adaptive thresholding computes the local mean luminance around each pixel window so ink is separated from local paper brightness regardless of lighting gradients.

Because printed text consists of thin high-frequency dark strokes while lighting shadows vary smoothly across large areas, a wide spatial box/morphological filter estimates the blank paper illumination map. Dividing each original pixel by its local background estimate normalizes the entire sheet to uniform white before contrast stretching.

Optical Character Recognition engines (like Tesseract, Google ML Kit, and Apple Vision) segment characters by locating connected contour boundaries. Removing paper texture noise, bleed-through from the reverse page, and lighting gradients prevents broken character strokes and false punctuation artifacts.

300 DPI (2,480 x 3,508 pixels for A4 or 2,550 x 3,300 pixels for US Letter) is the universal standard for OCR and legal archival. A 12 MP smartphone camera (4,000 x 3,000 pixels) filling the frame with a single page easily exceeds 340 DPI.

Never. All pixel manipulation runs locally on your CPU/GPU via the browser's HTML5 2D Canvas ImageData API. You can disconnect from the internet before dropping your file and the scanner will work identically.