Abstract

Image segmentation for partitioning an image into a set of meaningful regions is a key step in color image analysis. One basic technique is a histogram thresholding method using 3-D color features of RGB signals. However, the performance of segmentation depends not only on its algorithm but also on the color features. We have developed a useful segmentation method by means of the three perceptual attributes of hue, lightness, and saturation. The algorithm is based on a recursive thresholding of the histograms. The Munsell color system is used as the perceptually uniform color space. We use direct mapping to assign observed RGB signals to Munsell attributes for each pixel quickly. We give a criterion for finding the most significant peak from all the histogram peaks. A set of subregions with uniform color is extracted, and the operation is repeated on the remaining data to generate a sequence of uniform color regions in an image. Some additional algorithms are also described.

© 1987 Optical Society of America