Ranked-Based Distance Metric Learning: An Application to Image Retrieval

摘要

We have designed a prototype Content-based image retrieval (CBIR) system, called Tattoo-ID, for tattoo image matching and retrieval. CBIR systems automatically determine the image content in the form of low-level image features to compute the similarity between two images, rather than relying on human-assigned (external) class labels. We have examined several key design issues related to building a prototype CBIR system for matching and retrieving tattoo images. Tattoos are imprints on the skin that are being increasingly used by forensics and law enforcement agencies for identifying victims and suspects. Our prototype and first of a kind system demonstrates that it is possible to apply CBIR to this application domain. Initial retrieval results show great promise and pave the way for continued large scale development in this area.