Content-Based Hierarchical Classification of Vacation Images

Grouping images into (semantically) meaningful categories using low-level visual features is a challenging and important problem in content-based image retrieval. Using binary Bayesian classifiers, we show how high-level concepts can be understood from low-level images under the constraint that the image does belong to one of the classes in question. Specifically, we consider the hierarchical classification of vacation images; at the highest level, images are classified into indoor-outdoor classes, outdoor images are further classified into city-landscape classes, and finally, a subset of landscape images is classified into sunset, forest, and mountain classes. We demonstrate that a small codebook (the optimal size of codebook is selected using MDL principle) extracted from a vector quantizer can be used to estimate the class-conditional densities of the observed features needed for the Bayesian methodology. The classifiers have been built on a database of vacation photographs. Our system achieved an accuracy of for indoor-outdoor classification, for city vs. landscape classification, for sunset vs. forest & mountain classification, and for forest vs. mountain classification. Our final goal is to combine multiple -class classifiers into a single hierarchical classifier. M. Figueiredo was partially supported by Nato Grant NATOCRG 960010 & Portuguese PRAXIS XXI program, grant no. 2/2.1/T. I. T./1580/95.