Definition of descriptors for semantic image interpretation

A lot of effort has been put into researching image interpretation, but there is still no universally accepted approach to map low-level feature into high level image semantic interpretation. In this paper, a method for continuous low-level features vector quantization is presented so as to define appropriate values for descriptive variables. The similarity among different concepts of the domain is examined and compared by using the measure of similarity which is based on the probabilistic model and the measure of distance. Also, an abstract image description vector suitable for image analysis is given.