The program helps find similar photos, for example, one original unedited photo on disks with Windows, Mac OS and Linux (example). Alternatively you can use Duplicate Finder to find multiple groups of similar images in a directory (image clustering).
This reverse image search allows to find resized pictures, photos after small color/spacial changes, and exact copies. Perceptual similarity is identified by comparison of brightness of image sub-regions.
JPG, PNG, BMP, WEBP, GIF (static and animated) of maximum size 20MB per image. Transparency is ignored in image comparison.
Read about this photo comparison algorithm and check out image similarity in Go (Github) for your company or project. The open-source Golang code implements the image comparison function. The code does not include the duplicate image finder (clustering algorithm).
Parameters are set to find reasonably high number of near-similar images with acceptable speed and low false positives. The idea is to offer ready-to-use solution for majority of typical cases without many customizable parameters.
Images are compared by perceptual similarity in terms of color and brightness across the whole image. Partial occlusions, e.g. watermarks, impact perceptual similarity. If the perceptual changes are small, the pictures are considered similar. Rotated and mirrored images are considered distinct.