Upscale Image (OpenCV)

@opencv1 VersionEnhance

Enlarges an Image by an integer factor using a classical CPU interpolation kernel (nearest, bilinear, bicubic, or Lanczos) and emits the resized Image; fast and detail-preserving but not generative.

How it fits

Typical backends

Cheaply enlarge a low-resolution feed and publish the result: a file source feeds frames, this component resizes them with a classical kernel, and the output is streamed over HTTP.

Upscale small text crops before OCR so the recognizer sees larger glyphs, then publish the recognized text as JSON.

Caveats

LimitationClassical interpolation resamples existing pixels and cannot invent detail; pair with a neural super-resolver when reconstructing genuinely lost high-frequency content matters.
Parameter interactionThe minimum-width and minimum-height floors gate the skip TOGETHER: upscaling is skipped only when every non-zero floor is already satisfied, so an image already at or above the target size is passed through untouched. A floor of zero removes the constraint on that dimension.
I/O contractThe factor is an integer multiplier; the output can be clamped to maximum dimensions after enlarging, and the result is a same-color-space Image.
CostInference runs entirely on CPU with no model download, so it is cheap and has no GPU requirement, making it a good default when generative quality is not needed.

Versions

430bac89latestdefault
Automated release
6/21/2026