Edge Detection Image Tool

How Edge Detection Works

This tool applies a convolution filter to each RGB channel independently. It uses a fixed 3x3 kernel for the selected method, repeats the nearest border pixel outside the canvas, and adds a bias of 128 to the filtered result. The alpha channel remains unchanged.

The available methods are:

• Roberts Cross – kernel [1,0,0;0,-1,0;0,0,0]
• Sobel-Feldman – kernel [-1,-2,-1;0,0,0;1,2,1]
• Laplacian – kernel [-1,-1,-1;-1,8,-1;-1,-1,-1]

Strength linearly blends the original and filtered bytes: 0% keeps the source, 50% is halfway, and 100% uses the full biased convolution.

Expected Output and Example

The result is a biased directional or second-derivative response, not necessarily a black-and-white outline. It does not combine horizontal and vertical Sobel magnitude, trace vector paths, or guarantee measurement-grade edges.

For a 3x3 grayscale ramp from 0 to 8, the center pixel at full filter strength becomes:

• Roberts: 124
• Sobel: 152
• Laplacian: 128

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

Practical Checklist

• 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.

Limitations and Errors

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 MB 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 tool adds a bias of 128 to the convolution result, so edges appear as bright or dark areas on a mid-gray background. This is not a binary outline but a gradient response.

Can I adjust the kernel size or threshold?

No, each method uses a fixed 3x3 kernel. There is no threshold or kernel-size control. You can only change the method and strength.

Does the tool combine horizontal and vertical Sobel edges?

No, it applies a single Sobel kernel. It does not compute the magnitude from both directions, so directional edges may be missed.