Image

Image Edge Detector

Detect edges with Sobel, Prewitt or Laplacian convolution, with gradient direction, threshold control and an explanation of what each operator responds to.

Last reviewed by the Radiatus Cloud team

Drop an image here, click to browse, or paste one
Everything happens in your browser. Nothing is uploaded anywhere.
Preview appears here.

Need this done properly for your business?

Radiatus delivers secure cloud, DevOps & compliance engineering.

Book a free consult

An edge is a place where brightness changes quickly

Edge detection finds where the image gradient is large, which corresponds to object boundaries, texture and anywhere the surface or lighting changes abruptly. It works by convolution: a small matrix slides over the image and each output pixel is a weighted sum of its neighbours. Different matrices respond to different things, and the choice determines what the result actually shows.

Sobel and Prewitt measure the gradient, Laplacian measures curvature

Sobel and Prewitt each apply two kernels, one horizontal and one vertical, and combine the results to give both the strength and the direction of the gradient at every pixel. Sobel weights the centre row more heavily, which smooths slightly and makes it less sensitive to noise than Prewitt. The Laplacian is a single kernel measuring the second derivative, so it responds to where the rate of change itself changes, producing thinner edges and considerably more noise.

Noise is the practical problem

Differentiation amplifies high frequencies, and noise is high frequency, so every edge detector amplifies whatever grain is in the image. Blurring slightly before detection is the standard remedy and is why the Canny detector, the usual production choice, begins with a Gaussian blur. Skipping that step gives an edge map full of speckle that no threshold cleans up satisfactorily.

Related tools

  • EXIF Viewer — Extract hidden EXIF data (Camera, GPS, Settings) from images.
  • Favicon Generator — Generate favicons and app icons at every required size from one source image, with the exact HTML to reference them.
  • Placeholder Image Generator — Generate placeholder images at any size with custom colours and label text. Produced in your browser as downloadable files, no external hotlinking.
  • Image Resizer — Resize images in your browser with aspect ratio control and format conversion. Files are processed locally and never uploaded.

Frequently Asked Questions

What is convolution?

Sliding a small matrix over the image and making each output pixel a weighted sum of its neighbours. Almost every image filter is a convolution; only the matrix differs.

Which operator should I use?

Sobel for general purposes, since it weights the centre row more heavily and is less sensitive to noise. Prewitt is simpler and noisier. Laplacian gives thinner edges and much more noise, and is mainly useful for sharpening.

Why is my edge map so noisy?

Because differentiation amplifies high frequencies and noise is high frequency. Blurring slightly before detection is the standard fix, and it is why the Canny detector begins with a Gaussian blur.

What does gradient direction tell me?

Which way the brightness is changing, which is perpendicular to the edge itself. It is what allows edges to be thinned to a single pixel and is the basis of shape and orientation analysis.

Is this the same as the Canny detector?

No. Canny adds a Gaussian blur, thins edges to one pixel by suppressing non-maximum gradients, and links them with two thresholds. It is the production standard and these operators are the gradient step inside it.

Privacy & Security

Everything runs in your browser; nothing is uploaded.

Data: None
Client-side-Side
Active
v1.0

How to Use

Upload an image to see its edges detected by convolution.

Disclaimer: This tool is provided "as is" without warranty of any kind. Results are for educational and utility purposes.