Image classification for content-based indexing

Grouping images into (semantically) meaningful categories using low-level v isual features is a challenging and important problem in content-based imag e retrieval. Using binary Bayesian classifiers, we attempt to capture high- level concepts from low-level image features under the constraint that the test image does belong to one of the classes. Specifically, we consider the hierarchical classification of vacation images; at the highest level, imag es are classified as indoor or outdoor; outdoor images are further classifi ed as city or landscape; finally, a subset of landscape images is classifie d into sunset, forest, and mountain classes. We demonstrate that a small ve ctor quantizer (whose optimal size is selected using a modified MDL criteri on) can be used to model the class-conditional densities of the features, r equired bg the Bayesian methodology. The classifiers have been designed and evaluated on a database of 6931 vacation photographs, Our system achieved a classification accuracy of 90.5% for indoor/outdoor, 95.3% for city/lands cape, 96.6% for sunset/forest & mountain, and 96% for forest/mountain class ification problems. We further develop a learning method to incrementally t rain the classifiers as additional data become available. We also show prel iminary results for feature reduction using clustering techniques. Our goal is to combine multiple two-class classifiers into a single hierarchical cl assifier.