A New Peak Selection Criterion Based on Minimizing the Classification Error

Abstract Automatic selection of the number of subimages is a classical and challenging problem of image segmentation. An efficient way to select the number of subimages is to determine the number of peaks from a histogram. In this paper, a new criterion for peak selection based on minimizing the classification error is presented. Its application to image quantization is also studied. The value of the criterion for each gray level, called strength, is computed based on minimizing the error probability and maximizing the Mahalanobis distance. The number of peaks can be selected either subjectively or objectively. The experimental results show that the peaks of the histograms are selected properly even when the histograms are extremely noisy.

[1]  Julius T. Tou,et al.  Pattern Recognition Principles , 1974 .

[2]  P.K Sahoo,et al.  A survey of thresholding techniques , 1988, Comput. Vis. Graph. Image Process..

[3]  Nikos Papamarkos,et al.  A New Approach for Multilevel Threshold Selection , 1994, CVGIP Graph. Model. Image Process..

[4]  S. Soltani,et al.  SURVEY A Survey of Thresholding Techniques , 1988 .

[5]  M. Ibrahim Sezan,et al.  A Peak Detection Algorithm and its Application to Histogram-Based Image Data Reduction , 1990, Comput. Vis. Graph. Image Process..

[6]  José Manuel Rebordão,et al.  An amplitude segmentation method based on the distribution function of an image , 1984, Comput. Vis. Graph. Image Process..

[7]  Du-Ming Tsai,et al.  A fast histogram-clustering approach for multi-level thresholding , 1992, Pattern Recognit. Lett..

[8]  Richard O. Duda,et al.  Pattern classification and scene analysis , 1974, A Wiley-Interscience publication.