基于主动式边界基元模型的多类目标识别方法

针对形状特征,提出了一种基于主动式边界基元模型的多类目标自动识别方法. 该方法以主动式边界基元为基础构建字典,可准确描述各类目标的形状结构, 不受尺度、旋转等变化的影响;然后,综合分析上下文信息进行概率学习,采用级联框架和Bootstrap动态采样训练最优边界分类器,实现目标的类别识别和位置定位,并可获取精确形状. 实验结果表明,该方法能有效提取多种类型和复杂结构的目标,具有较强的实用价值.

Multi-categorical object recognition using method based on active contour basis model

A new multi-categorical object recognition method based on the active contour basis model is proposed. The method builds a class-specific codebook of active contour bases, which is robust to scale variation and pose changes. Probabilistic learning by analyzing contextual information is performed using cascaded frame and boot strap dynamic sampling. A classifier is trained to determine the object categories and exact regions. Experimental results demonstrate that the proposed method achieves high efficiency in extracting manifold and complicated objects.