Image access and data mining: An approach

Abstract

In this paper, we propose an approach that discovers automatically visual relations in order to make more powerful the image access. The visual relationships are discovered automatically from images. They are statistical rules in the form of a → b which means: if the visual feature "a" is true in an image then the visual feature "b" is true in the same image with a precision value. The rules concern symbols that are extracted from image numerical features. The transformation of image numerical features into image symbolic features needs a visual feature book in which each book feature is the gravity center of similar features. The approach presents the clustering algorithm that creates the feature book.