A multimodal search tool for artworks

We search for information online every day, whether for things to buy, the news, technical information or academic research. We might imagine that this is a problem solved, until we remember the frustration of being unable to locate exactly what we want among the noise of everything else. Our project solves this in a very specific environment - fine art.

We provide clients with authenticity and attribution studies, where we need to quickly find comparison artworks for evaluation and, increasingly, to train our expert AIs. Search is key to this.

Searches work by pre-indexing the source data for rapid retrieval. However, results are highly dependent on what is being indexed. For artworks we need to find artefacts that are closely comparable across multiple factors, some of which are visual (things that 'look' similar) and others factual, such as artist, place, or date.

Our project uses multiple indexing measures, including some based on how connoisseurs evaluate paintings, such as palette and balance. This achieves greater precision and accuracy while allowing us to modify search parameters according to need. Our aim is to deploy this approach of multiple indexing measures to other creative sectors like practising artists, publishers and so forth.