You finish a phone photo of a printed contract, and the gray paper refuses to separate from the ink. A single luminance cutoff forces every pixel to either white or black, which is ugly on portraits yet perfect when you want a stencil, a fax-style mask, or a quick OCR-friendly plate. The workspace below keeps pixels on your device, so an NDA screenshot or a hospital form never transits our servers.
Adaptive compares each pixel with a local mean (15×15 neighborhood, minus a constant). The slider is ignored while adaptive is active.
Office scanners often deliver 300 dpi grayscale TIFFs that already look acceptable on screen. Thresholding still helps when you need a one-bit fax mask or when you paste the graphic into a slide template that demands pure black strokes. Start from the native scan, not a JPEG re-save, because each lossy cycle adds mosquito noise around letters.
When forms include checkboxes, a slightly lower global cutoff keeps pencil graphite visible without filling entire squares. When forms use blue ink, the green channel sometimes separates better than luminance; this page does not expose per-channel thresholds, so we recommend fixing separation upstream or accepting a single-channel compromise.
Batch workflows belong on the desktop. This interface targets one honest preview at a time: you see source and binary together (or flip between them on a narrow phone), you hear the slider snap through values, you download when the halos look tolerable. For redaction chains where you must hide faces after binarizing text, chain photo censor only after you trust the mask edges.
We convert RGB to luminance with the usual ITU-style weights (roughly 0.299 red, 0.587 green, 0.114 blue), then compare against either your global value or the local mean. Alpha copies forward unchanged, so transparent PNGs stay transparent where the source allowed.
Nothing uploads. The FileReader API pulls bytes into memory, canvas writes pixels, and toDataURL builds the PNG when you hit save. If you need to prove tonal spread before you commit to a cutoff, open color distribution on a duplicate export and read the histogram story there.
Receipt on a wooden table: Global threshold at 150 blows out the varnish grain while keeping the thermal print readable. Adaptive pulls the same text up when the wood darkens toward the frame.
Signature on blue ink: Inverted mode yields a white glyph on black, which pastes cleanly into slide decks without fiddling blend modes.
Low-resolution logo PNG: Threshold exaggerates anti-alias halos. We recommend smoothing edges in a vector tool after export, or stepping back to smart optimizer if file size drove the harsh compression.
Thresholding is not edge detection. If you need gradients turned into lines, edge detection operators answer a different question. Thresholding is also not dithering: you will not retrieve midtones once they collapse to black or white.
Very large images stress mobile GPUs; adaptive mode multiplies work. For a 24 megapixel dump, consider downsampling in a desktop editor first, then fine-tune here.
Short answers tied to this page, not generic photo help.
Adaptive computes a different cutoff per pixel from local brightness. The global slider would fight those per-pixel decisions, so the UI disables while adaptive runs.
PNG, because binary art needs lossless channels. JPEG would reintroduce gray mush around edges you just removed.
Many engines preprocess internally. Some prefer grayscale. Run an A/B test on your pipeline; if OCR worsens, send the grayscale sibling from black and white conversion instead.
The browser decodes to RGBA first. Embedded CMYK profiles map before threshold math, which is fine for quick tests but not a substitute for print-calibrated workflows.
The luminance histogram probably sits in a narrow band. Lift shadows or tame highlights elsewhere, then return. Extreme single-tone frames have no meaningful cutoff.
