Human-Centered Content-Based Image Retrieval

In this chapter, general color concepts are introduced, which will be used in the remainder of this thesis, for color analysis, colorful texture analysis, shape extraction, and CBIR. First, the color histogram is described: a discrete function that quantizes the distribution of colors of an image. Next, several color spaces and their color quantization schemes are discussed. In addition, an alternative view on color is presented: 11 color categories. A brief history is sketched of a century of research on these color categories. Consecutively, the Sapir-Whorf hypothesis, the work of Brown and Lenneberg, and the theory of Berlin and Kay, are discussed. 2.1 The color histogram Color is the sensation caused by light as it interacts with our eyes and brain. The perception of color is greatly influenced by nearby colors in the visual scene. The human eye contains two types of visual receptors: rods and cones. The rods are responsive to faint light and therefore, sensitive to small variations in luminance. The cones are more active in bright light and are responsible for color vision. Cones in the human eye can be divided in three categories, sensitive to long, middle, and short wavelength stimuli. Roughly these divisions give use to the sensations of red, green, and blue. The use of color in image processing is motivated by two principal factors. First, color is a powerful descriptor that facilitates object identification and extraction from a scene. Second, humans can discern thousands of color shades and intensities, compared to about only two dozen shades of gray [95]. In this chapter, general color concepts, as used in this thesis, will be introduced. We will start with a description of the color histogram. Next, color quantization will be explained, followed by the description of several color spaces and their quantization methods. In addition, the research conducted in the last century toward an alternative view on color is presented: 11 color categories. We end this chapter, with the introduction of the distance measures that have been applied. 2.1 The color histogram The color histogram is a method for describing the color content of an image, it counts the number of occurrences of each color in an image [321]. The color histogram of an image is rotation, translation, and scale-invariant; therefore, it is very suitable for color-based CBIR: content-based image retrieval using solely global color features of images. However, the main drawback of using the color histogram for CBIR is that it only uses color information, texture and shape-properties are not taken into account. This may lead to unexpected errors; for example, a CBIR engine using the color histogram as a feature is not able to distinguish between a red cup, a red plate, a red flower, and a red car as is illustrated in Figure 2.1. Many alternative methods have been proposed in the literature. They include color moments [217, 285], color constants [85, 321], color signatures [142], color tuple histograms [104], color coherent vectors [105], color correlograms [88], local color regions [142], and blobs [51]. These methods are concerned with optimizing color matching techniques on a spatial level; i.e., utilizing the spatial relations between pixels, in relation to their colors. However, they disregard the basic issue of intuitive color coding. In other words, the way the engine is processing color, is not related to human color processing. In our opinion, prior to exploring these techniques, the issue of color coding (or categorization) should be stressed; e.g., as can be done with the color histogram.