Automatic Image Annotation Based on Low-Level Features and Classification of the Statistical Classes

Abstract

This work is devoted to the problem of automatic image annotation. This problem consists in assigning words of a natural language to an arbitrary image by analyzing textural characteristics (low-level features) of images without any other addtional information. It can help to extract intellectual information from images and to organize searching procedures in a huge image base according to a textual query. We propose the general annotation scheme based on the statistical classes and their classification. This scheme consists in the following. First we derive the low-level features of images that can be presented by histograms. After that we represent these histograms by statistical classes and compute secondary features based on introduced inclusion measures of statistical classes. The automatic annotation is produced by aggregating secondary features using linear decision functions.