Chapter 15: Color Image Segmentation

With contributions by Gertjan J. Burghouts

In this chapter, we consider the invariant assessment of color and texture in combination with applications in image segmentation and material classification. 1 For texture segmentation, we consider the work on Gabor filters [283] and Gaussian derivative filters as the most important [284, 285]. For the modeling of materials, the mapping of image features onto a codebook of feature representatives receives extensive treatment. For reason of generality and simplicity, filterbank outputs are commonly used as features. These methods are often referred to as texton-based methods [286, 287], or nowadays bag-of-word approaches. The combination of color and texture has attracted attention in the recent literature. In Mirmehdi and Petrou [288], color-textured images are roughly segmented based on a spatial color model [289]. The assumption underlying their approach implies that texture can be characterized by its color histogram over a region. The drawback here is that the spatial structure of the texture is not considered since only first-order statistics, the histogram, is taken into account. Thai et al. [290] propose measuring color–texture by embedding the Gabor filters into an opponent color representation. The method provides a useful structural representation for color–texture.

We show the extension of the Gaussian color model presented in Chapter 6 to the domain of texture by extending the Gaussian ...

Get Color in Computer Vision: Fundamentals and Applications now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.