A growing number of art institutions are making their collections available online but search and discovery of these collections relies on searching by attributes, which has limited scope for hands-on exploration.
We have developed a content based search system by coupling compressed image representation and approximate nearest neighbour (ANN) search, and use it as a building block for creating an engaging method for visual discovery. The system comprises of two parts: image feature representation and ANN search.
The system uses a Deep Convolutional Neural Network for compressed image representation and approximate nearest neighbour search to find and display visually similar images. The system runs in real time, making it an engaging and playful way of discovering content in a rich archive. Embracing the external environment adds an experiential dimension, which drives the desire to try the application in different environments.
This project is ongoing.