Detect Edges in an Image
Pick an image and a method — Canny or Sobel edges are traced in your browser, nothing is uploaded.
About this tool
Edge detection reduces a photo to the boundaries where brightness changes sharply — the outlines an object-recognition pipeline, a CNC/laser cutter or a vector tracer works from. Upload an image, pick a method, and the edge map is computed locally by a WebAssembly build of ffmpeg. Your file never leaves the browser and nothing is uploaded to a server.
The three methods
| Method | What it does | Best for |
|---|---|---|
canny | Gaussian smoothing → Sobel gradients → non-maximum suppression (thinning) → hysteresis between the two thresholds. White 1-pixel edges on black. | Clean, connected outlines; tracing; anything you will threshold later |
sobel | Raw gradient magnitude only. Grey edges whose brightness tracks contrast strength — no thinning, no thresholds. | Soft or blurry photos, texture/relief maps, engraving depth |
colormix | Canny detection, but the edges are painted over the original picture instead of replacing it. | An inked/cartoon look that keeps the photo readable |
Worked example
Take a 640×480 photo of a building, facade.jpg, and run the Clean outline preset —
method = canny, low = 0.2, high = 0.5, blur = 1.
The result is facade.png: a 640×480 black image with thin white lines along the window
frames, roofline and door edges, and the brick texture gone (the 1-pixel blur removed it
before detection, and the raised thresholds dropped what was left). Tick Invert and the
same run returns black lines on a white background — a printable coloring-page version.
Lower the thresholds to 0.03 / 0.1 instead and the brickwork comes back as a dense mesh
of edges.
Choosing thresholds
Both thresholds are fractions of full brightness (0–1), not 0–255 values, so the same numbers work on any image. Hysteresis means: a gradient stronger than high always starts an edge, and the edge is then followed through neighbouring pixels as long as they stay above low. That is why one threshold gives broken dashes and two give continuous lines.
- Start with the defaults (0.078 / 0.196 — the classic 20/255 and 50/255).
- Too much noise? Raise both, or add 1–2 px of blur first.
- Missing faint edges? Lower low to about a third of high.
- A high:low ratio between 2:1 and 3:1 is the usual recommendation.
Limits and edge cases
- Maximum input size is 8 MB; larger files are rejected rather than silently downscaled.
- Input can be PNG, JPEG, WebP, GIF, BMP or TIFF — anything the browser build of ffmpeg decodes. Output is PNG (default), JPEG or WebP.
- Animated GIF/WebP inputs are reduced to their first frame; this tool returns a still image, not an animation.
lowmust not exceedhigh; the tool reports that instead of producing an empty frame.lowandhighare ignored whenmethod = sobel— that operator has no thresholds.- Transparency is discarded:
cannyandsobelconvert to grayscale first, so a PNG with an alpha channel is detected against its composited pixels. - Very large blur values (towards the 10 px cap) will erase all detail and return an almost empty edge map — that is the filter working, not a failure.
FAQ
What is the difference between Canny and Sobel edge detection?
Sobel computes the brightness gradient at every pixel and stops there, so its output is a soft grey magnitude image where thick, fuzzy bands mark strong contrast. Canny starts with the same Sobel gradients but adds three steps: smoothing to suppress noise, non-maximum suppression that thins each ridge down to a single pixel, and hysteresis thresholding that keeps only edges connected to a strong one. The result is a crisp binary line drawing. Use Canny when you want outlines to trace, Sobel when you want to see how strong the contrast is.
Why is my edge map almost empty (or almost solid white)?
Both symptoms are threshold problems. An empty map means high is above nearly every
gradient in the picture — lower high, and lower low to roughly a third of it. A
solid-white map means the thresholds are so low that sensor noise, film grain and JPEG
blocking all register as edges — raise them, and set blur to 1–2 pixels so the noise is
smoothed away before detection. Low-contrast or very dark photos usually need both a lower
threshold pair and a brightness/contrast pass beforehand.
Can I get black lines on a white background for printing?
Yes — tick Invert. The detector always produces white edges on black internally; invert flips the result, which is the form you want for printing, coloring pages, laser engraving and most vector-tracing tools. The Coloring page preset combines invert with slightly raised thresholds and a blur pass so only the major outlines survive.
Does this upload my image anywhere?
No. The page loads a WebAssembly build of ffmpeg and runs the whole filter chain inside the browser tab, so the picture and the edge map both stay on your machine. The first run downloads the ffmpeg engine (a few MB, cached afterwards); after that the tool works offline. The same detection is available from the command line, where files are read locally too.
What image formats and sizes are supported?
Input: PNG, JPEG, WebP, GIF, BMP and TIFF, up to 8 MB. Output: PNG (the default, and the right choice for high-contrast line art), JPEG or WebP. JPEG compresses thin white lines badly — you will see grey ringing along every edge — so prefer PNG unless file size matters more than fidelity. Animated inputs contribute only their first frame.
Why is edge detection useful?
It is the first stage of most classical computer-vision pipelines: shape and contour detection, document and card boundary finding, OCR pre-processing, and measuring objects in a scene. Outside vision, edge maps feed CNC routers, laser engravers and vinyl cutters, serve as the starting point for vector tracing, and are used in image compression research and in art/illustration workflows where a photo needs to become a line drawing.
Developer & Automation Access
Run it from the terminal
Same engine as this page, headless — via the gizza CLI:
gizza tool edge-detection 'url=https://example.com/input' 'method=canny' 'low=0.078' 'high=0.196' 'blur=0' 'invert=true' 'format=png'New to the CLI? Get gizza →
Open it by URL
Pre-fill and auto-run this tool with query parameters — the names match the API/CLI:
https://gizza.ai/tools/edge-detection/?url=https://example.com/input&method=canny&low=0.078&high=0.196&blur=0&invert=true&format=pngMachine-readable descriptor: tool.json — title + parameters JSON Schema for agents.