Abstract: Combining the two essential characteristics of the image, the gradient and the gray level, a threshold segmentation approach using maximum entropy with the gradient boundary control was proposed. In this approach, a boundary gradient controlling function was defined to quantify the richness of the detail information in the images. The local maximums of this function indicated a possible threshold set for the image segmentation. Within this set, a best threshold could be selected by using maximum entropy principle to get the binary image. The experimental results show that the regions segmented by this method can contain more semantics of the images for the ample detail information reserved in it. Moreover, this method is also with noise immunity in some degrees.

Key words: image segmentation, threshold, maximum entropy, gradient, detail information

摘要： 结合梯度和灰度这两种图像的本质特征，提出一种基于边界梯度控制的最大熵阈值分割方法。该方法首先定义了一种边界梯度控制函数来定量分析图像中细节信息的丰富程度，通过该函数的局部极大值确定可能的分割阈值的集合，然后根据最大熵原理在该集合中选取最优阈值，最终实现图像的二值化分割。实验结果表明该方法的分割结果由于保留了丰富的细节信息，能够更好地体现图像语义，且该方法亦具有一定的抗噪性。

关键词: 图像分割, 阈值, 最大熵, 梯度, 细节信息