Relation of binary image complexity to gray‐scaled image thresholding

Conventional gray‐scaled image thresholding consists in finding the threshold according to the bimodality of a gray‐level histogram of an input image. However, few discussions have been made on the thresholding problem from the viewpoint of the optimality of the results. As far as image perception (human perception) is concerned, we naturally see the image as simply as possible. Taking such situation into account, the threshold for image thresholding operation should be determined on the same basis. This paper presents a principle in which the optimality of the thresholding process is evaluated according to the complexity change of the binarized picture against the change of the threshold value. Specifically, we show that the complexity curves are categorized into either unimodal or multimodal. For the multimodal case, we show that a good thresholding can be obtained by choosing a minimal point on the curve as the threshold. For a unimodal case we introduce another strategy which depends on the local multimodality of the complexity curve.