INDI - Techniques for Intelligent Navigation in Digital Image Databases

In the last decade the quantity of digital images grew rapidly with the development of new technologies like digital cameras, scanners or camcorders. This causes the need for structured data organisation to search efficiently in digital image datasets. The traditional method to save digital imagery in the filesystem suffers from the individual organisation and expensive search strategy which is realized by manual browsing through the filesystem. Motivated by the database technology digital images are organised in database systems. Based on the textual annotations of the images the stored dataset can be searched by using the assigned key words. Such systems are affected adversely by the subjective image perception and the time consuming manual textual annotation. Modern image database systems join traditional database technology with different elements of computer vision and pattern recognition. Such content-based retrieval (CBR) systems offer the possibility to search in image datasets by using automatically extracted image features. In this project we develop a system for intelligent and interactive navigation in digital image databases using content-based image features. On the one hand we will offer the user the possibility to interact with the database in an intuitive way. Therefore the frontend is one of the important developing tasks. And the keyboard of a computer should not the only instrument to interact with the database. Furthermore speech and gestures should amount the comfort. On the other hand we analyse various approaches for computing content-based image features as well as suitable distance measures. During the usage of the database INDI should learn as much as possible about the user requests. Therefore we implement adaptive procedures to refine the queries. A relevance feedback based on a user assessment handles this task.