Abstract

Traditional search engines such as Google are only capable of indexing and searching text, yet about 73% of the data on the web consists of images and video. Meanwhile, consumers and businesses are accumulating millions of digital photos, yet they lack efficient tools to browse, organise, search, and retrieve them.

To address these challenges, my colleagues and I have developed a content based image retrieval methodology that enables computers to automatically index and search images by visual content. The system features a range of image processing and analysis modules which can automatically recognise semantic image content (e.g. beaches, sky, faces etc.). It allows ordinary users to intuitively search for pictures using text queries consisting of keywords or short natural language sentences, which are then linked to image content using an ontology of visual concepts.

However, with an estimated 15 billion images available on the internet today, there is great need for technologies to provide the scalable computing resources required to process such vast quantities of data. With the help of funding from the UK STFC and in collaboration with the Cambridge University eScience Centre, I am developing techniques to scale my image analysis approach to millions of digital pictures using distributed grid processing technology. This was done by porting the software to the middleware architecture deployed on the UK particle physics grid, which consists of several 10s of thousands of computers. We are also furthering middleware and user interface development to make it easier for other non-physics applications from academia or industry to make use of grid computing.