On the statistical and computational performance of image thresholding and determination of class number

This paper proposed a new and fast method for bilevel as well as multi-level image thresholding. Taking the derivative of image between-class variance with respect to gray levels develops the proposed method. For bilevel thresholding, a nonlinear equation is derived to solve for an optimal threshold. For multi-level thresholding, a set of nonlinear equations is derived to solve for a set of optimal thresholds. Two parameters are introduced to automatically determine the number of classes for image classification. One parameter is designated for classifying images with obvious histogram modes. The other is more appropriate for use for images without obvious histogram modes. Computational efficiency of the proposed method and Otsu's method (1979) is discussed. Statistic analysis of performance of the proposed method versus Baysian classifier is conducted. Included are two examples to illustrate the feasibility and superiority on speed of computation of the proposed method.