A Comparative Evaluation of Statistical Image Segmentation Techniques.
SOUTHEASTERN MASSACHUSETTS UNIV NORTH DARTMOUTH DEPT OF ELECTRICAL AND COMPUT ER ENGINEERING
Pagination or Media Count:
Statistical image segmentation refers to the computer-oriented procedures that partition the image into meaningful parts by using the statistical pattern recognition techniques. Although most image segmentation works have been nonstatistical in nature, there is now strong interest in the use of the supervised and the unsupervised classification techniques for image segmentation. In this paper, a critical comparison is made on the supervised image segmentation techniques including the Fishers linear discriminant, the autoregressive moving-average modelling, the maximum likelihood region estimation, and the maximum a posteriori region estimation, as well as on the unsupervised image segmentation techniques including the cluster analysis, the estimation-theory based method, histogram directed segmentation techniques, and the decision-directed method using the conditional population mixture model. Some computer results are presented. The fundamental issues in the statistical image segmentation and the related topics are also reviewed. Author