THRESHOLDING OF DIGITAL IMAGES USING 2-DIMENSIONAL ENTROPIES

Thresholding is an important form of image segmentation and is a first step in the processing of images for many applications. The selection of suitable thresholds is ideally an automatic process, requiring the use of some criterion on which to base the selection. One such criterion is the maximization of the information theoretic entropy of the resulting background and object probability distributions. Most processes using this concept have made use of the one-dimensional (1D) grey-level histogram of the image. In an effort to use more of the information available in the image, the present approach evaluates two-dimensional (2D) entropies based on the 2D (grey-level/local average grey-level) "histogram" or scatterplot. The 2D threshold vector that maximizes both background and object class entropies is selected.