An unlikely team of scholars started work on something called Ontology based search from a leafy Cambridge base in 1995. Now this team is supported by a group of local investors and, likes to think of itself as the Google of image search. Imense claims to be two years ahead of its rivals in visual search technology. Now Google and Microsoft are watching. I talked to Tony Rowland, head of sales and Stuart Cox, agency partner manager to get through the tough academic terms and develop a better understanding of the benefits of the system and the latest innovations in the field.
David Sinclair, with a background at Oracle, Olivetti and AT&T started to work on visual search in 1995. At AT&T he met Christopher Town in 2001. Shortly after they decided to join forces and start a company. Chris brought new theories on “ontology (linguistic and visual) based visual information processing”.
The company that was founded in 2004 was called Cambridge Ontology with cash and government grants. This funding allowed them to complete the model which took about 3 years. Ten years research became the foundation of what the company does today and in 2007 Cambridge Ontology became Imense. The two co-founders secured another round of investment in only three months with a number of companies interested. The 5 investors that were chosen are from a technology background and a good fit for the company. Their cash allowed the roll out of the commercial model and the resulting product was launched in August 2008.
The team of two has since grown to eight full time and five part-time staff with most of the team focussed on technology. This team has built a portfolio of two search and two keyword products. Today the founders David and Chris are still very much involved in the business. They both have an academic background with Chris being very well known for receiving both a dissertation award for his final work and an audience with the queen. The chairman Henrik brings commercial acumen. Together with the rest of the group this creates an eclectic team of both technology, sales and creative thinking who’s hobbies include rock-climbing, fencing and even singing in a choir.
Web 3.0 was the first product based on visual information processing. The goal was to find and serve up images that had now keywords. This flagship product is supplied on a harddrive if purchased and will act like a plugin enabling the clients’ website to use the ontology search. It analyses images and will recognise objects on the basis of a number of elements in the picture. This includes face recognition and even the age of the subject.
Duplicate and similar search was the second product to come from Cambridge. This was initially for publishers who had trouble finding and tracking images through their system. It is now also used in the Stock Photography industry with AGE, IPN stock and Capture integrating it into their services.
Imense worked with photographers, agencies and websites to find out more about their search behaviour. While Imense stores over 500 features of an image only few of them where important for clients in finding similar images. These were colour, structure and texture. This hierarchy of elements leads to a high accuracy in locating similars.
At Imense, the team consideres this similar search basic technology and level one in a ten layer system that they are building. An interesting element of similar search where companies can search on composition of an image which allows searching companies for on-brand images. Other companies providing a similar product are Picscout, LTU and the startup Imprezzeo.
Automated Image Annotation (Autotagger) is a service where an image is keyworded by a computer. Initially 40-60% accurate this is now up to 90% as a result of working with 12-15 initial categories like travel, celebrities and wildlife. After these categories are set the computer takes the visual information and generates a comprehensive list of keywords. The goal is to reduce operational cost for users of the system, in particular for the celebrity imagery that comes to market in large quantities.
This product is for Stock agencies and publishers. It will take under one second to process and keyword an image before the client gets back a CSV file or the data embedded with the image. Imense is working on teaching the system about concepts as well to further improve the accuracy and is working to roll out the system per category over the course of the year. While Imense understands a human keyworder will still be necessary for conceptual search it thinks it can get very close with its system.
Finally there is a keyword application called Annotator. This tool helps keyworders and photographers on their way by reducing keywording time from four days to only one. Annotator is webbased. Files can be uploaded and then keyworded. This starting point is created by visual search technology which creates a number of base keywords to build on. It is aimed at a keyworder working on location and accelerating their work.
One of the keys is the Hotlist functionality where users can create their own associative keyword lists. Similar to shared lightboxes online, these can then be shared with other users creating shared knowledge and experience. Autotagger is priced at about $0,09 an image. It has a minimum order of 5.000 images due to the time it takes to set up the system. Annotator, which is aimed at individual photographer has a fixed price of $145,-.
Imense also has its own image-site. No transactions take place here; instead clients are referred directly to the affiliate’s site. While it works with 3.0 search it’s not merely a demo site but also generates some revenue. Imense now speaks to bigger companies like Google and Yahoo regularly and keeps them posted on the latest research and products. Its clients include Stock agencies but also museums, newspapers, digital asset management providers. On top of that facial recognition technology can be used in the security business like border control.
Will Autotagger run out of images to keyword? At Imense they don’t think so as it can play a role in tagging billions of images that are online. For the next 18 months though the Stock Media industry will be the number one priority. For the short term the team is focussing on improvements and delivering autotagger to clients. As nobody else keywords images automatically this is going to be hard work. The agency platform is another focus. Annotator will be launched at PDN Photoexpo and there will be more news at PACA later in the year. This small company is working hard to change the way Stock Media clients find images and while they are based in a location that is not immediately associated with exciting technology start-ups it has established a strong fan-base. Their location has also not prevented them from being on the radar of some Global companies that are keeping a close eye on the latest news from Cambridge.
Image: www.sxc.hu / Bridge / Anthony Shapley