Edge Detection Image Tool

Detection methods

Sobel–Feldman combines the signed horizontal and vertical 3 × 3 gradients as sqrt(Gx² + Gy²). Roberts Cross combines two signed diagonal differences in a 2 × 2 neighborhood in the same way. Laplacian uses the absolute eight-neighbor second derivative.

RGB channels are processed separately. Samples outside the image repeat the nearest border pixel; alpha and pixel dimensions stay unchanged. No offset of 128 is added.

Roberts uses a fixed 2 × 2 neighborhood; Sobel and the eight-neighbor Laplacian use fixed 3 × 3 neighborhoods. Kernel size and threshold are not adjustable; choose the method and strength.

Strength

Strength ranges from 0% to 100%, defaults to 50% and changes in steps of 0.01. The edge magnitude is clamped and rounded to a byte before blending with the original RGB byte; the blend is rounded again to the nearest byte, with ties to even. At 0% the decoded source RGB is retained.

Worked example

For the 3 × 3 grayscale image [0,1,2;3,4,5;6,7,8], the source center is 4. At 100% strength, the center values are Roberts 4, Sobel 25 and Laplacian 0; at the default 50%, they are 4, 14 and 2 respectively.

Use and limitations

The result may contain colored edges because RGB channels are processed independently; it is not a thresholded binary outline. Canny, threshold adjustment, vector tracing, object recognition and measurement-grade accuracy are not provided. Compare the preview with the source.

The available methods are:

This illustrates how each method responds to intensity changes. Always review the preview against the original before download.

• Use a clear, high-contrast image for best results.
• Start with the default Sobel method and 50% strength.
• Adjust strength to balance edge visibility and noise.
• Compare the preview with the original to verify edges.
• Remember that output is a PNG with the same dimensions as the input.

This tool processes one image at a time. It does not offer Canny edge detection, threshold control, or kernel-size adjustment. The output is not a vector path or a measurement-grade edge map. If the image exceeds 25 MiB (26 214 400 B) or 24,000,000 pixels, it may fail to load. Ensure the file is a browser-decodable format.

Frequently asked questions

Why does my result look gray instead of black and white?

The result may contain colored edges because RGB channels are processed independently; it is not a thresholded binary outline.

Can I adjust the kernel size or threshold?

Roberts uses a fixed 2 × 2 neighborhood; Sobel and the eight-neighbor Laplacian use fixed 3 × 3 neighborhoods. Kernel size and threshold are not adjustable; choose the method and strength.

Does the tool combine horizontal and vertical Sobel edges?

Yes. Sobel–Feldman combines both signed horizontal and vertical gradients as sqrt(Gx² + Gy²).