Comparing color edge detection and segmentation methods

The authors report the results of a controlled comparative study of four proposed methods for extracting boundaries in static images using color information. The first two techniques are, more properly speaking, edge-detection methods. These include the adaptation of the Hueckel operator to color images by R. Nevatia (1977) and the DOOG (difference of offset Gaussians) operator proposed by R. Young (1985, 1986). The latter two methods are examples of region-based segmentation. These include the adaptation of R. Ohlander et al. (1978), region splitting by Y. Ohta et al. (1980), and the fuzzy c-means segmentation and edge-detection methods of T.L. Huntsberger et al. (1985). The methods are compared using both simulated, controlled images as well as a variety of natural scene images. Images with various levels of randomly additive noise provide some indication of the degradation of performance of these methods under noisy conditions