Categorization is a work to classify the category of the data. Nowadays, demand on categorization is increasing enormously due to the availability of large databases and performance requirements, such as speed, accuracy, and cost. In particular, image categorization is a very useful and important technique in the image retrieval or in filtering system. In many of the emerging researches and applications, it is clear that any single approach for the image categorization is not optimal and thus other approaches adapting multiple methods have to be used. Consequently, combining several methods and classifiers is now commonly and practically used in the category classification.
In this paper, we propose an automatic image categorization method using MPEG-7 techniques. Generally, the design of an image categorization essentially involves the following three aspects: 1) image acquisition, 2) image representation, and 3) category decision-making. The categorization belongs to the decision-making part. In our method, the images are categorized by content-based description using MPEG-7. All similarity distances at each category are measured using multiple MPEG-7 descriptors. In addition, a matching technique for combining similarity distances using multiple MPEG-7 descriptors is newly addressed. As further work, we devise a method to apply the ordinal measure to categorization. Because most of errors have similar pattern, we try to find the pattern by considering the correlation with other categories. So it will be more reduce the possibility that the query image is not correctly classified.
Experiments are performed with some query images, which are classified by fourteen categories. Experimental results show that the proposed category classification method is superior to that with single descriptor.