摘要
针对现实世界中外观不一,而形态上有模式可循的对象,提出了一种基于模板库的自动对象分割方法。采用模板描述对象形态,构建模板库作为对象的先验知识,将未知的匹配模板视为隐变量,采用由按位编码方式组成的复合模板与图像特征进行匹配确定隐变量的取值空间;然后,对隐变量求积分得到图像中对象的边缘概率。最后,求解边缘概率所对应的能量函数,实现基于模板库的自动对象分割。实验结果验证了论文方法的正确性和有效性。
For some objects with different appearance but following definite mode on contour in the real-world,a template-library-based automatic object segmentation approach is provided.First of all,template is used to describe the object contour,and build a template library as the prior knowledge of objects.Afterward,the unknown matched template is regarded as a hidden variable,and by matching the bit-encoded composited template and image features,the value space of variable is determined.Then the integral operation is utilized to get the marginal probability of the labeling given the image data.Finally the energy function is solved corresponding to the marginal probability to achieve template-library-based automatic object segmentation.Experimental results demonstrate that the approach is accurate and effective.
出处
《舰船电子工程》
2014年第5期91-96,共6页
Ship Electronic Engineering
关键词
对象分割
图割
模板
能量函数
object segmentation
graph cut
template
energy function