The Near-Duplicate Detection model is used to identify images that are so called duplicates or near-duplicates. It can be used across the following use-cases:
The duplicate detection works with all types of images, both natural (such as standard photos) and artificial: drawings, logos, abstract content... It will detect duplicates of an image across a wide range of transformations and modifications, many of which are typically used to try to evade or circumvent duplicate detection. Here are examples:
Changes to the dimensions of the image, to the resolution, to the image format and encoding (such as PNG, JPG, WEBP...)
The deduplication is robust to the addition overlays such embedded text items, watermarks, logos etc
The deduplication model is crop-resistant, meaning that images that have been cropped, or have an added border/frame will be detected
Color modifications such as transforming the image to black-and-white, sepia, changing the saturation, brightness, contrast, hue and other color manipulations
The deduplication model will detect a duplicate even if the image has been rotated
Compressing, stretching (changing the aspect ratio), or flipping the image
Other types of image edits where specific parts of the image are modified, enlarged, transformed or have their colors changed
Choose your use-case to learn how to use the Image Deduplication model:
Blacklist images and prevent them from (re)appearing on your site or app. For instance copyrighted images, illegal images, previously removed images.
Detect spammy and unwanted behaviors. Prevent users from submitting the same image multiple times, and from submitting other users' photos.