摘要
存贮目标所有的视点图像,建立完备的特征库,或者提取能够抵抗视点变化的不变特征,是三维目标识别的常用方法。这两种方案都存在不足:要么特征库规模庞大,识别过程计算量大,识别效率低;要么难以找到鲁棒的识别特征。结合两种方案研究了基于聚类技术建立三维舰船模型特征库的方法。利用仿射传播聚类方法无需事先指定聚类中心的优点,将其应用于两型舰船模型的视点空间聚类。通过提取视点图像的Hu矩特征,进行了仿真实验,给出了聚类结果的有效性分析。
Creating a sufficient feature base for all of the viewpoints or extracting an invariable feature of different viewpoints are the common methods for 3I） target recognition. They share the shortcomings of big size feature base, enormous computation and low recognition rate. It is almost impossible to find the recognition feature of robustness. The method for establishing the 31） naval vessel models feature base with clustering technology was studied. The affinity propagation （AP）clustering algorithm requires no initializing cluster centers and it is suitable to cluster view points space. Experiment of AP was done using computer based on the Hu mo- ments, and the clustering results validied the method by Silhouette index.
出处
《应用光学》
CAS
CSCD
北大核心
2012年第2期260-264,共5页
Journal of Applied Optics
关键词
目标识别
视点空间
无监督聚类
仿射传播聚类
target recognition
view points space
unsupervised clustering
affinity propagationclustering
