Online Batch Image Edge Detection 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.

Batch edge detection accepts up to 30 selected images and applies one shared Detection method and Strength setting to every loaded image.

Generate creates one browser preview per loaded image. Each result is encoded as PNG and named from the source base name plus -edge.png. Download all starts separate downloads about 180 ms apart; it does not create a ZIP or preserve the source format.

Per image: ≤25 MiB and ≤24,000,000 pixels. Per batch: ≤100 MiB and ≤40,000,000 pixels. Selecting over 30 files or any invalid image rejects the whole batch.