Improving Image Database Summarization by Automatic Image Feature Selection Using GA

Abstract

The problem of image feature selection is addressed, within the framework of the image classification and image database summarization applications. The existing image classification methods, as well as the image database summarization method proposed by Stejić et al. (2001), require manual selection of image features used for the classification, which makes these methods difficult to design, domain-dependent, and non-optimal. We extend the original image database summarization method by the automatic feature selection procedure, based on genetic algorithm. The proposed method is evaluated through comparison with the original one, on two image databases, each with 1000 photographs, partitioned into 10 semantic categories. The proposed automatic feature selection procedure improves the performance over 11% in average. The proposed method enables the user an easy access to the image database contents, by bridging the gap between a large number of images in a database, and a typically small number of semantic categories those images represent. Furthermore, through the proposed automatic feature selection procedure, the method is able to adapt to the diverse content and dynamic nature of the image databases, typical for the Internet.