Painting recognition from wearable cameras

Overview & Synopsis

Are smart glasses the new high-tech device that will guide you through a museum? In this report, we describe a system that runs "on device" on Google Glass and retrieves the painting you're looking at among a set of paintings in a database. We perform an experimental comparison of the accuracy and speed of different feature detectors and descriptors on a realistic dataset of paintings from Musée du Louvre. Based on this analysis we design an algorithm for fast and accurate image matching using bags of binary features. We develop an application that runs directly on Google Glass without sending images to external servers for processing and recognizes a query painting in a database of 100 paintings in one second. The code is available online.