Image-based Similarity Search
With the image similarity search, you can explore an inventory of several million automatically recognised images. The images include both individual illustrations (photos, graphics, paintings) and illustrations in books or manuscripts. The search criterion applied relates to the similarity of image motifs according to purely outward characteristics such as colours, textures or contrasts.
You can search for people, animals, plants, landscapes, coats of arms, book covers, miniatures or initials. Give it a go and search for "digital twins".
Instructions for use
You can use a picture from this thematic selection page to access the initial screen. However, you can also use any other image from bavarikon's extensive collection as a search template. To do this, activate the option "Zeige Bildähnlichkeitssegmente" (Show image recognition segments, eye icon), when viewing a certain image or an illustrated page of a book. Existing image segments are marked with a blue frame and can be used as a search template; one click triggers the recognition search. The search results can be opened and inspected in the viewer; they can also be used as a template for further recognition searches. Please note that the image segments are selected automatically. Incorrect or incomplete sections may occasionally occur, or segments with irrelevant content may be detected. Take your time: you will probably need several successive search attempts until you start getting useful results.
Search with upload
Upload your own image as a template for an image search
- Format JPG/JPEG or PNG only! The recommended resolution is 150 to 300 ppi. Limit is 1500 KB per file.
- Images can be colored or grayscale (black & white), but should not be too pale. To avoid negative effects on the search result, use motifs without borders, without text and without much white space.
- Please note that a search with uploaded files takes longer than usual (about 15 to 20 seconds). The uploaded file will be deleted after the search.
No similar images found.