基于等势线特征的图象特征域抽取研究

Research on Extracting Equipotential-line-similar Region from Image

图象识别检索的关键步骤是提取出稳定特征,引申了物理学中等势线的概念到图象特征区域中,并提出了两种方法:图象重排法和基于极小值灰度点的区域生长法来抽取这类图象特征区域。理论分析表明图象重排法在算法时空间复杂度上相对较优于基于极小值灰度点的区域生长法。实验表明,图象重排法可取得较高的运行处理速度,部分抽取特征域甚至具备语义特征。

In this paper,the region with the property that all pixel inside the region have either higher or lower intensity than all pixels on its outer boundary is taken as the feature region according to the definition of equipotential line used in Physics,and two new methods to extract such regions based on image-resorting processing and region-growing processing is presented.Theoretical analysis shows that time complexity of the method based on image-resorting processing is better than the other one.Results of experiment indicate that the method based on image-resorting processing get more stable efficiency,and some region extracted can be taken as meaningful components of object in images.