Medical Image Database Kernel with a NN Selection Driven Image Retrieval Algorithm
Vasile Cornita, Rodica Strungaru, Sever Pasca, Mihaela Ungureanu and Florin Perianu
This paper presents an image database system kernel with image retrieval capabilities based on semantic information associated with image, target image or combined search, in which the image retrieval algorithm selection is carried out via a Neural Network (NN) based on previous experience. The Image Database System Kernel (DBMSK), which is composed of two software applications: the DBMSK Server and Client, uses typical three-layer architecture. For retrieval by target image, we investigate three histogram based search methods in RGB and HSV color spaces: histogram intersection, histogram Euclidian distance and histogram quadratic distance. Because a histogram characterize an image by color distribution, information about image shape and location being discarded, better search result are obtained when image has text semantic information attached and search takes into account this semantic information.
Keywords: Database system kernel, content based image retrieval algorithms, neural networks, object oriented programming, artificial intelligence, histogram intersection, euclidian distance, quadratic distance.