物体形状特征的准确提取往往受到图像获取、目标分割等过程中各种因素的影响,因而需要构造一种鲁棒的形状表现来保有其本质特征。揭示弹性二次曲线(EQW)模型中能量权重因子和样条重叠度因子与形状几何特性间的关系,提出一种自适应保持几何特征的EQW形状表现模型,并将该模型与交互式Live Wire算法相结合,应用至图像分割中,对于各种模态图像中的目标进行分割实验。直观上,本文方法能够在去除噪声同时较好地保持目标形状的显著特征;量化结果方面,交互式分割误差保持在较小稳定范围内,引入EQW模型带来的时间开销亦在可接受范围内。

Shape feature extraction from an image usually suffers from effects like imaging noise and segmentation errors. A robust shape description algorithm is therefore a prerequisite. In this paper, relations between performances on elastic quadratic wire (EQW) based shape representation model and the model parameters are revealed first. By taking advantage of these relations, we focus on building a new EQW model that adaptively preserves geometrical features. Then the proposed EQW model is further embedded into the Live Wire algorithm. In experiments, the improved Live Wire algorithm is tested on medical and remote sensing image segmentation. Qualitatively, the results show that the proposed model has better performances on extracting object boundaries while avoiding meaningless morphology caused by noisy or incomplete data. Quantitatively, the segmentation errors and the temporal cost of the improved Live Wire are both acceptable.