Color Segmentation Based on Separate Anisotropic
Diffusion of Chromatic and Achromatic Channels

Luca Lucchese ^a,b and Sanjit K. Mitra ^a
a Dept. of Electrical and Computer Eng.
University of California at Santa Barbara
Santa Barbara, CA 93106
^b Dipartimento di Elettronica e Informatica
Universit a diPadova, Italy
Via G. Gradenigo 6/A, 35131 Padova
phone: (805) 893 8312, fax: (805) 893 3262
e-mail: fluca,mitrag@iplab.ece.ucsb.edu
 

Abstract

This paper presents a new technique for segmenting images only on the basis of color infor-
mation. It is shown how segmentation can bene t from splitting color signals into chromatic
and achromatic channels and separately smoothing them through anisotropic di usion. Oper-
atively, this is accomplished through two independent di usion rocesses: one involves only the
chromatic information, conveniently embeddedin a complex function, while the other a ects the
lightness information. The results of the two di usions are separately segmented by k-means
clustering techniques and their combination yields the nal image partition into homogeneous
regions. Some experimental results are reported in the paper which verify the e ectiveness of the
proposed technique.