Title:	Local vs. Global Histogram-Based Color Image Clustering
Authors:	Bongani Malinga, Daniela Raicu, Jacob Furst
Abstract:	In this paper, we present two image clustering techniques to automatically group color images that correlate with semantic concepts. This work goes towards satisfying the ever growing need for techniques that are capable of automatically generating semantic concepts for images from their visual features. We present two techniques and evaluate their relative performances based on the perceptual similarity among images that are clustered together. The first technique is based on the localized histogram information while the second approach uses global histogram information to characterize the images. Experiments using a 2100 image database are presented to show the relative effectiveness of the presented systems. Preliminary results show that the local histogram approach gives better clustering results and its characterization of images is more closely related to the way in which humans perceive images.
Full Paper:	[pdf]