Digital image databases have seen an enormous growth over the last few years. However, since many image collections are poorly indexed or annotated, there is a great need for developing automated, content-based methods that would help users to retrieve images from these databases. In recent times, a lot of attention has been paid to the management of an overwhelming accumulation of rich digital images to support various search strategies. In order to improve the traditional text-based or SQL (Structured Query Language)- based database searches, research has been focused on efficient access to large image databases by the contents of images, such as color, shape, and texture. Content-based image retrieval (CBIR) has become an important research topic that covers a large number of domains like image processing, computer vision, very large databases, and human computer interaction (Smeulders, Worring, Santini, Gupta & Jain, 2000). Several content-based image retrieval systems and methods have recently been developed. QBIC (Query By Image Content) is one of the first image retrieval systems developed at IBM (Niblack et al., 1993). Color, texture, and shape features are combined to represent each image in this system. The VisualSeek system, developed at the Columbia University, is an image retrieval system based on visual features (Chang, Smith, Mandis & Benitez, 1997). The NeTra system is a prototype image retrieval system, which uses color, texture, shape, and spatial location information as features to retrieve similar images (Ma & Manjunath, 1997). Some of the other popular CBIR systems are MARS (Ortega et al., 1998), Blobworld (Carson, Thomas, Belongie, Hellerstein & Malik, 1999), PicToSeek (Gevers & Smeulders, 2000), and SIMPLIcity (Wang, Li & Wiederhold, 2001). An analysis of these systems reveals that all of them give a lot of importance on the image color for retrieval. In fact, color is always considered to be an important attribute, not only in content-based image retrieval systems, but also in a number of other applications like segmentation and video shot analysis. In color-based image retrieval, there are primarily two methods: one based on color layout (Smith & Chang, 1996) and the other based on color histogram (Swain & Ballard, 1991; Wang, 2001). In the color layout approach, two images are matched by their exact color distribution. This means that two images are considered close if they not only have similar color content, but also if they have similar color in approximately the same positions. In the second approach, each image is represented by its color histogram. A histogram is a vector whose components represent a count of the number of pixels having similar colors in the image. Thus, a color histogram may be considered to be a signature extracted from a complete image. Color histograms extracted from different images are indexed and stored in a database. During retrieval, the histogram of a query image is compared with the histogram of each database image using a standard distance metric like the Euclidean distance or the Manhattan distance. Since color histogram is a global feature of an image, the approaches based on color histogram are invariant to translation and rotation, and scale invariant with normalization. Color histograms may be generated using properties of the different color spaces like RGB (Red, Green, and Blue), HSV (Hue, Saturation, and Intensity Value), and others. In this article, we give an overview of the different histogram generation methods using the HSV color space. We first present a brief background of the HSV color space and its characteristics, followed by the histogram generation techniques for various applications.
Content Based Image Retrieval: A Review
