# Background Remover

> Remove plain backgrounds from product photos and logos (colour-based, no AI) → transparent PNG. Makes the background of an image transparent by colour: it estimates the background colour from the image border and removes it with an adjustable tolerance, or removes one colour everywhere (chroma key / green screen). Works well for product photos, logos and scans on plain backgrounds; it does NOT understand people or complex scenes, because it uses no AI model.

Part of usefulweb.site (https://usefulweb.site/). Results are deterministic calculations from the inputs and the listed assumptions; the API returns exactly what the web page shows.

## Use it
- Human interface: https://backgroundremover.usefulweb.site/ (interactive tool at https://backgroundremover.usefulweb.site/tool)
- JSON API: POST https://backgroundremover.usefulweb.site/api/v1/calculate with a JSON object of inputs (schema: https://backgroundremover.usefulweb.site/api/v1/tool)
- OpenAPI: https://backgroundremover.usefulweb.site/openapi.json
- MCP (Streamable HTTP): https://backgroundremover.usefulweb.site/mcp - tool `background_remover`

## Inputs
- file (file, required): Image - JPG, PNG, WebP, AVIF, GIF, TIFF or HEIC, max 50 MB. Best results: the subject on a plain, evenly lit background.
- mode (select): Method
- color (text): Background colour - Hex colour, e.g. #ffffff or #00ff00. Required for the two colour methods; ignored by the automatic method.
- fuzz (number): Colour tolerance - How different a pixel may be from the background colour and still be removed. Raise it for shadows or JPEG noise; lower it if parts of the subject disappear.
- cleanup (integer): Edge clean-up - Removes isolated specks and fills small holes in the mask (morphological open/close). 0 = off.
- feather (number): Feather edges - Softens the cut-out edge. 0 = hard edge.

## Method
- This is NOT an AI background remover. It decides per pixel by colour distance only, so it works for a subject on a plain, evenly coloured background (product shots, logos, scans, green screen) and fails on busy scenes, hair, or subjects with the same colour as the background.
- Automatic mode shrinks the image to 200×200 px and takes the per-channel median of a 3-pixel border ring as the background colour. The share of border pixels within the tolerance ("border uniformity") predicts how well the method will work.
- Edge-connected modes add a 1-pixel frame in the background colour and flood-fill from it with ImageMagick (-fuzz, -floodfill), so only background regions touching the edge are removed and same-coloured areas inside the subject stay. Chroma-key mode removes every pixel close to the colour, wherever it is.
- The resulting mask is cleaned with a morphological open (removes specks) and close (fills pin holes) using a disk of the chosen radius, optionally blurred to feather the edge, and used as the alpha channel of a PNG.
- Existing transparency is flattened onto white first. Colour tolerance is a root-mean-square distance over R, G and B relative to the full 0–255 range.

## Limits
- Colour-based removal is approximate. Check the edges in the preview; for portraits or complex scenes use an AI tool or manual masking.