摘要
传统Otsu及其大多数改进的算法只将类间方差作为最优阈值的唯一依据,使得对直方图分布不同图像的分割效果差异很大。针对该问题,提出一种新的阈值分割方法。建立一个耦合对象相似度模型,考虑对象的各种属性以及属性之间的关系,以捕获对象间的关联关系。采用耦合对象相似度替代传统Otsu算法中的类间方差,将所选阈值划分出的每个类分别看作耦合对象相似度模型中的一个对象,每个类都有类的概率和灰度均值这2个属性,计算类间相似度,并在类间相似度最小时取得最优阈值。实验结果表明,与传统Otsu、二维Otsu、大熵分割算法相比,该算法能提高刻画类间差异的精确度和图像的分割精度。
Since the Otsu method and most of its improved methods take between-class variance as the foundation of picking threshold,which causes great difference in segmentation for image with different histogram distribution, a new threshold segmentation algorithm is proposed in this paper. Firstly, a model of Coupled Object Similarity (COS) is introduced, which can take both the relationship of the various attributes of the object itself and the relationship between the properties into account,and can capture the relationships between the objects. Secondly,between-class variance in the Otsu method is replaced by COS to pick threshold. Each class distinguished by the selected threshold is regarded as an object in the model of COS. Each class has two attributes, the probability of class and gray mean. Similarity between classes is calculated, and the optimal threshold value is obtained according to the minimum of similarity between classes. Experimental results show that compared with Otsu algorithm, two-dimensional Otsu algorithm and maximum entropy algorithm,the algorithm can measure the difference of classes at a higher accuracy and obtains better segmentation results.
出处
《计算机工程》
CAS
CSCD
北大核心
2016年第10期255-260,共6页
Computer Engineering
关键词
图像分割
方差
耦合对象相似度
类间相似度
最优阈值
image segmentation
variance
Coupled Object Similarity ( COS )
between-class similarity
optimal threshold