On Image Classification: To Retrieve Images from Major Web Sites

Lakshman Jayaratne, Athula Ginige, Zhuhan Jiang


Grouping images into semantically meaningful categories using basic low-level visual features is a challenging and important problem in content-based image retrieval. The enormity and diversity of the visual contents on the Web images adds another dimension to this challenging task. Moreover, the retrieval of Web images cannot be easily achieved with images of other than trivial collections, and therefore one needs to put more cognitive load on the users. Based on the groupings, effective indices can however be built for an image database. In this paper, we show how a specific high-level classification problem can be solved from relatively basic low-level visual primitives geared for the particular classes. We have developed a procedure to qualitatively measure the saliency of a feature towards a classification problem based on the discrimination power of the HSV color histogram. We found that the HSV color histogram, mainly the hue component, has the most discriminative power for the classification problem of our interest. A k-means classifier is then used for the classification, which results in an accuracy of 91.8% when evaluated on an image database of 2,738 Web images. The images are classified as full faces, natural sceneries, events and city images. Our final goal is to use this classification knowledge to enhance the performance of content-based image retrievals by filtering out images from irrelevant classes during the matching.