Iterative Approximation of Empirical Grey-Level Distributions for Precise Segmentation of Multimodal Images
- PDF / 1,898,200 Bytes
- 15 Pages / 600 x 792 pts Page_size
- 8 Downloads / 175 Views
Iterative Approximation of Empirical Grey-Level Distributions for Precise Segmentation of Multimodal Images Ayman El-Baz CVIP Laboratory, Department of Electrical and Computer Engineering, University of Louisville, Louisville, KY 40292, USA Email: [email protected]
Aly A. Farag CVIP Laboratory, Department of Electrical and Computer Engineering, University of Louisville, Louisville, KY 40292, USA Email: [email protected]
Georgy Gimel’farb Centre for Image Technology and Robotics (CITR), Department of Computer Science, University of Auckland, Tamaki Campus, Auckland 1000, New Zealand Email: [email protected] Received 29 December 2003; Revised 5 December 2004 A new algorithm for segmenting a multimodal grey-scale image is proposed. The image is described as a sample of a joint Gibbs random field of region labels and grey levels. To initialize the model, a mixed multimodal empirical grey-level distribution is approximated with linear combinations of Gaussians, one combination per region. Bayesian decisions involving expectation maximization and genetic optimization techniques are used to sequentially estimate and refine parameters of the model, including the number of Gaussians for each region. The final estimates are more accurate than with conventional normal mixture models and result in more adequate region borders in the image. Experiments show that the proposed technique segments complex multimodal medical images of different types more accurately than several other known algorithms. Keywords and phrases: density estimation, image segmentation, expectation maximization.
1.
INTRODUCTION
A large number of image segmentation methods based on estimating marginal probability densities of signals and separating their dominant modes have been developed and tested for the last three decades (see [1, 2, 3, 4, 5] to cite a few). However many important applications such as medical image analysis or industrial vision still encounter difficulties in separating practically meaningful continuous or disjoint objects, even when signal densities are distinct to the point where their mixture becomes strongly multimodal. The basic issue is the accuracy of region borders, which are usually essential for correct interpretation of the objects. Typically, the fine separation of the signal modes to specify the region borders is obtained by intersecting tails of the signal distributions for the adjacent objects. Therefore, it is the tails that have to be precisely estimated in order to separate, for example, a darker object from a brighter background. One of the practical problems that inspired our approach is to accurately
detect lungs in a spiral CT chest slice so that their borders closely match those outlined by a radiologist. Because there always exists an overlap between the signal ranges for the different objects, the precise segmentation has to account for spatial distributions of the signals, too. Markov-Gibbs random field models show considerable promise in spatial image analysis [6, 7, 8, 9, 10, 11, 12, 1
Data Loading...
