Trade.Information

Future of image searching

Looking for images on the internet can be a frustrating business. Whether you want the perfect sunset over the sea or the London skyline by night, you’re dependent on people to describe the images on their web pages. Now Imense Ltd, a high-tech Cambridge start-up, has announced new investment to help them become the ‘Google’ of image searching, using their revolutionary technology. To test their software, they’ve made an unexpected partnership with a group of particle physicists using a massive computer Grid.
Professor Keith Mason, Chief Executive of the Science and Technology Facilities Council (STFC) describes how this happened “We actively encourage the researchers we fund to consider the wider applications of the work they do. In this case, computing problems that had to be addressed for particle physics can also be used to solve other challenges with large amounts of data. The Council’s Knowledge Exchange Service put the two teams together and provided modest funding to start them off – the new investment attracted by Imense represents a ten-fold return on the initial development funds.”
Images and video make up over 70% of the digital data available on the Internet, an estimated 15 billion images, but traditional search engines can’t index this information directly, instead relying on text descriptions entered by hand. Imense’s key innovation is a new form of image retrieval that automatically analyses images in terms of their content, without the need for human generated captions. They have also developed a powerful query language that lets people search for the images they need.
Dr David Sinclair, one of the founders of Imense Ltd, explains, “We built a prototype of our new image analysis and search technology, but simply weren’t able to test our software on sufficiently large numbers of photos. We knew we could search tens of thousands of pictures, but couldn't afford to try it on hundreds of thousands or millions of images. This made it difficult for Imense to get the investment we needed to develop a commercially viable product. That’s where our partnership with the particle physics Grid came in.”
Spread across 17 sites, the UK particle physics Grid (GridPP) has been built to analyse the petabytes of data expected from Europe’s newest particle accelerator, the Large Hadron Collider. But its 8000 computers have also been shared with other researchers, from geophysicists to biologists. Last year, Sinclair attended a meeting arranged by STFC about Grid opportunities for industry, and realized that Grid technology could be the answer to Imense’s problem. Image analysis is a naturally parallel process which fits perfectly with the capabilities of the Grid used by STFC scientists to process data in particle physics.
Professor Andy Parker, Director of the eScience Centre, University of Cambridge, led the particle physics team working with Imense, “Our team helped Imense develop their software to run on the Grid using a tool called Ganga, and supported them as they analysed three million images. We also dealt with issues such as security and working with Grid managers at other universities, who were very helpful. It went very smoothly and was fascinating to see the company start-up process in action.”
Imense have now reaped the rewards of their Grid experience, with an investment of more than £500,000 to help them bring a product or service to market in the coming months. Dr Sinclair says that the Grid played a major role in this, “Our work with the Grid has let us demonstrate that our software can handle millions of images, at a time when we were a small company and couldn’t supply the computing power needed ourselves. This in turn impressed the investors we spoke to, and led to funding for our company.” Imense plans to use the open source Grid technologies from the particle physics domain in its commercial product.
Alex Efimov led the brokering work for STFC’s Knowledge Exchange Service and companies wishing to know more should contact him on the number below.

News Inside News:

The PIPSS Knowledge Transfer Scheme
PIPSS is a knowledge transfer scheme that supports the development of effective, long term collaborations between UK Universities, CERN, ESO (European Southern Observatory), ESA (European Space Agency), UK industry and research sector organisations, with the aim to:

Promote co-ordinated technology development within the Council programme and with other partners;
Encourage researchers to be aware of the possibilities for exploitation;
Raise awareness in industry and other research sectors of the technological strengths and opportunities afforded by the Council's science;
Encourage collaboration between UK companies and the research community;
This will ensure the maximum benefit to UK industry, through transfer of technologies and skills developed from the Council's research programme to broader market areas. Also encouraging a two-way transfer of skills and knowledge between the Council's supported scientists and researchers in UK industry and research organisations representing other sectors.

The overall objective of PIPSS is to facilitate the transfer of technology developed in pursuit of the Council's science programme. In other words, to transfer Science and Technology Facilities Council developed technologies to other academic disciplines and/or industrial applications

About Cambridge Ontology Ltd-
Cambridge Ontology Ltd has developed plain text driven content based image search technology. The company was founded in 2004 by two leading image retrieval scientists based in Cambridge, England.
Cambridge Ontology's technology makes content-based retrieval of images as easy and powerful as the search for text documents on the Internet. Our technology combines state-of-the-art computer vision with machine learning, natural language processing, and information retrieval methods to give an incomparable search experience.
The system automatically labels semantic visual content in images. Our cutting edge computer vision and visual classifiers allow the system to recognise a wide range of visual categories (e.g. grass, sky, sunset, beach etc.), as well as to spot and interpret faces.
Our technology allows a user to type a plain text query to search large image collections for images with matching visual content.

Details of the technology-
The Problem
Digital search technology is a multi-billion industry. Over the last years image search has grown by 91%, over 4 times the rate of web search for documents. But although over 73% of the data on the internet consists of images and video rather than text, current search technology used by Google treats images as black boxes and relies on captions or document context to index them. Many consumers are accumulating thousands of personal images, yet they lack efficient tools to browse, organise, search, and retrieve them.

Solution
Our key innovation allows users to search “inside the picture”, i.e. search over the actual content of an image. Content indexing of images is done completely automatically using state-of-the-art computer vision and image processing methods. Unlike other image search solutions, our system does not rely on image annotations or metadata, and does not require an initial example image or sketch. People can simply type a few keywords or optionally a more complex query using a standard browser interface. Our software is able to automatically interpret the meaning of the query and match it to relevant images purely on the basis of their appearance.

Opportunities
The success of businesses such as Flickr, MySpace and YouTube shows that user generated content is an enormous driver of consumer attention. Our technology offers a sophisticated yet completely automated way of making visual content searchable by content without any need for manual tagging while still retaining the ability to incorporate contextual information. We are thus not only in a position to offer superior image search compared to companies like Google and Yahoo, but also to make vast quantities of currently un-indexable user generated media content searchable first time.

Release link: http://www.camtology.com/