The representation of shape for retrieval of pictures by semantic means

Content-based image retrieval (CBIR) lets the user search an image database by means of image features rather than keywords. Although techniques for searching the database are quite advanced, the success of a CBIR system is still limited because few methods exists that close the gap between the semantics of the request and the features into which the request is to be translated. Progress can be expected, if knowledge about rules of human vision in the search and recognition of objects and object features in pictures are used for defining an image representation. It must be possible to extract this representation and it must be capable of integratuing semantic information from user feedback during search of the image database.