Machine Learning: Improved Photo Tagging with Relevance Feedback

Background

Camera phones represent a new class of networked media capture devices that combines: media capture; programmable processing using open standards and APIs; wireless networking; rich user interaction modalities; user metadata; and personal information management functions. The usage of media capture devices is undergoing transformation from the traditional camera-to-desktop-to-network pipeline to an integrated mobile media experience. Management and tagging of photos captured with a camera phone is therefore a natural development.

Relevance feedback has proven to be effective for improving retrieval accuracy in most retrieval models. It has proven to be worthwhile in a wide variety of settings, both when actual user feedback is available, and when the user feedback is implicit

The Thesis

The goal of this thesis is to extend a photo annotation system with relevance feedback. It involves development of algorithms, deployment as a web service and evaluation.

Using RDF/RDFS for structured metadata has the advantage of facilitating interoperability between disparate systems and supporting inferencing capabilities. Further we aim to provide an Android application for testing and evaluation.

Who we are

Multimedia Technology is a department within Ericsson Research where research about different media types such as audio, images and video is carried out. Multimedia applications and protocols technology is a subgroup, and here we deal with media transport and analysis. The thesis work will be carried out in Kista, Stockholm.