Content-Based Image Retrieval – A Words and Pictures Approach

The ever-increasing interest in digital image retrieval has led to research into ways of reducing and if possible eliminating the cost of traditional image annotation. This report describes the development of an automatic image annotation and contentbased image retrieval system. The process documented in this paper has an initial training phase in which the associations between words and images are learnt, with these associations then used to automatically annotate new images with no annotation. The report covers the design, implementation and evaluation of a system that functions as described above, with the conclusion that the developed system was considered successful and will provide a base for future development with a wide scope for the system’s improvement.