2 维对称交叉熵图像阈值分割.docVIP

2维对称交叉熵图像阈值分割 吴一全 张晓杰 （南京航空航天大学 信息科学与技术学院 南京 210016） 摘 要 现有阈值分割方法中交叉熵不满足对称性，通过运用对称交叉熵描述出2维对称交叉熵阈值选取公式快速递推算法将运算转换到两个1维空间上。实验结果表明，方法图像分割 阈值对称交叉熵 2维直方图TN911.73 文献标识码：A Two-dimensional Symmetric Cross-Entropy Image Thresholding WU Yi-quan, ZHANG Xiao-jie (College of Information Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 210016) Abstract The cross-entropy in the existing thresholding methods does not satisfy the symmetricity of distance measure. And the computation speed of the algorithms can be further improved. Thus an image threshold selection method based on decomposition and two-dimensional symmetric cross-entropy is proposed in this paper. Firstly, the difference between the segmented image and the original one is measured by the symmetric cross-entropy. The threshold selection formulae are derived based on the one-dimensional and two-dimensional symmetric cross-entropy, respectively. A two-dimensional fast recursive algorithm is given, which makes the computation complexity reduced to O(L2) from O(L4) of full search. Then the computation of two-dimensional symmetric cross-entropy is converted into two one-dimensional spaces and its computation complexity is further reduced to O(L). The experimental results show that, compared with the existing threshold selection method based on two-dimensional nonsymmetric cross-entropy, the proposed method has stronger anti-noise and the processing time is significantly reduced. It is a more effective threshold selection method based on two-dimensional cross-entropy. Keywords image segmentation, threshold selection, symmetric cross-entropy, two-dimensional histogram, recursive algorithm, decomposition0 引 言 图像分割是图像处理的基本技术，也是图像分析与识别的关键步骤之一。其中，阈值分割方法处理有效，实现简单，广泛应用于目标检测、指纹与印章鉴定、汽车牌照与字符识别、表面缺陷视觉检查等实际领域中。图像阈值分割的关键是如何选取阈值，以快速实现最佳分割效果。针对这一问题国内外学者进行了广泛的研究，提出了多种阈值选取方法[1]-[6]。这些方法首先根据1维或2维直方图及其区域划分方式，得到的阈值选取公式，由于交叉熵（相对熵、有向散度）描述了两个概率分布之间的差异程度，Li[7]和Brink[8]在图像阈值分割中先后引入了最小交叉熵准则，[9]-