Comparison of gray-scale thresholding algorithms applied to coronary arteriograms

An algorithm, called the iterative trinary classification (ITC) algorithm, is presented for image segmentation. Two gray-scale thresholds are used to define three pixel classes (object, background, and undecided). On the basis of a learning process, the two thresholds are iteratively modified until the number of undecided pixels drops to a certain level, the stopping criterion. The performance of the ITC algorithm on synthetic and human coronary angiograms is compared with that of a relation algorithm and a scattering based algorithm. The ITC algorithm is computationally efficient and provides the best segmentation performance. With its strong potential for speedup by exploiting parallel processing the algorithm should be useful in various computationally intensive procedures for the analysis of coronary angiograms