关键词: 图像分割, 多阈值, 谱聚类, 相似度, 一维直方图

Abstract: The thresholding is an important form of image segmentation and is used in many applications that involve image processing and object recognition. Thus, how to acquire a threshold of image segmentation is crucial. A novel multilevel thresholding algorithm is presented, which regards gray level L and corresponding probability P of 1D histogram of the image as points （L, P） in two-dimensional plane, and uses a new similarity function to define the similarity between any two points to construct the similarity matrix based on gray level, then uses the spectral clustering algorithm （Ncut） to classify the points, and the image thresholding is determined by the classification result. The similarity matrices are based on the gray levels of an image, rather than the commonly used image pixels. Therefore, the proposed algorithm occupies much smaller storage space and requires much lower computational costs. At the same time, this algorithm has the superior performance that is single-threshold and multi-threshold for image segmentation, compared to existing thresholding algorithms.

Key words: image segmentation, multilevel thresholding, spectral clustering, similarity, one-dimensional histogram