---

Image segmentation is the process of locating regions in an image that correspond to surfaces or objects in the scene. Segmentation is often an important first step in scene description or object recognition. In general, segmentation is a difficult problem because image measurements corresponding to a single region in the scene can have significant variation due to effects such as shading, highlights, nonuniform illumination, sensor noise, or texture. The use of accurate image models is therefore essential for accurate segmentation. In reference [1], we show from a careful analysis of reflection models that surfaces can be segmented independent of surface orientation or illumination direction. In reference [2], the analysis of reflection models is used to derive a segmentation algorithm that integrates local and global image information to account for image color variation due to changes in surface orientation and highlights. In reference [3], we use a new class of random field models for color texture to develop a segmentation algorithm for processing color images of natural outdoor scenes.

[1] G. Healey

``Using Color for Geometry Insensitive Segmentation,'' Journal of the Optical Society of America A, 6(6):920-937, June 1989. abstract

[2] G. Healey

``Segmenting Images using Normalized Color,'' IEEE Transactions on Systems, Man, and Cybernetics, 22(1):64-73, Jan/Feb 1992. abstract

[3] D. Panjwani and G. Healey

``Markov Random Field Models for Unsupervised Segmentation of Textured Color Images,'' IEEE Transactions on Pattern Analysis and Machine Intelligence, 17(10):939-954, October 1995. abstract

[4] G. Healey and L. Wang

``Three-Dimensional Surface Segmentation using Multicolored Illumination,'' Optical Engineering, 37(5):1553-1562, May 1998. abstract

Back to the ICVL homepage

Last modified: 28 July 1998