1/1999 - 12/2000
In the two-year Countess project, researchers have developed solutions for content-based image retrieval. The search platform prototype developed by MediaTeam researchers can be used to search for pictures in digital databases on the basis of their content. There are two search methods: the user can search for images according to selected example images, or he/she can draw or outline a model image with the mouse. The search criteria include colour, texture, shape, and content concept area.
A generally usable method for semi-automatic recognition of content concept areas has been implemented in the project. The method allows image retrieval systems to utilise information about image content. Since the tool can be trained to understand certain content concepts (e.g. sky, grass, and human skin), it can be utilised in a variety of image retrieval applications. The tool helps for example image archivists to add semantic information to an image database.
Until today, it has been possible to search for example the Internet for information with the help of one or more search entries. However, problems have resulted for example from the fact that the abstract and general visual content of an image is difficult to describe with search terms. It is also evident that people interpret images in different ways. Novel, more efficient image retrieval systems and search robots are more than welcome, since digital images are more and more commonly used.
Technological development has lead to the possibility to easily produce, process, transfer, and store even large numbers of images. For example different types of satellites, surveillance cameras, medical photographing, scientific experiments, fingerprint and face identification devices, as well as home electronics produce immense quantities of images.
Content-based image retrieval technologies can be utilised especially by those industries, which use large image archives. Image retrieval applications are used for instance by press agencies as well as news agencies, advertising agencies, architectural firms, engineering offices, copyright authorities, criminal investigators, military forces, medical experts, geologists, and even individual people for example to control their digital photo archives or to find images in the Internet.
A usable retrieval system has to respond to queries in a reasonable time. The system developed by our researchers has proved to function well with a database of ten thousand images in total. The system allows the user to give feedback of the success of his/her search. On the basis of the feedback, the search engine automatically adjusts the search parameters, and this leads to better search results.
The methods used in this research include image and texture analysis as well as image recognition methods. The design of the image retrieval system has been based on the modularity, expandability, and dispersibility of the system, as well as interfaces, performance speed, search ability, and search precision. The research problem is challenging, and this is why no major commercial breakthrough has yet been made in the field of content-based image retrieval systems.
The Countess project is a part of MediaTeamís research programme concentrating on content-based retrieval systems, which includes three currently running projects, one funded by the National Technology Agency and two funded by the Academy of Finland. The programme is aimed at finding out how multimedia and individual media types (e.g. text, image, sound, video) can be utilised on the basis of their content.
Online demos of CBIR demonstration systems
Online web demo of the developed CMRS (Content-based Multimedia Retrieval System) architecture (limited CBIR functionality)
Financiers and Business Partners
Acta Systems (now Yomi)
Research Program Affiliation
Usix - User-Oriented Information Technology 1999 - 2002