改进离散曲线演化的形状匹配算法

摘要

为了在形状匹配的过程中提高形状特征对边界噪声和图像变形的鲁棒性,同时兼顾形状匹配算法的检索精度和运算效率,提出一种基于离散曲线演化的形状匹配算法。首先利用改进的离散曲线演化算法对形状轮廓进行特征提取,获得具有重要视觉部件的轮廓；然后计算该轮廓的形状上下文特征描述符；最后利用动态规划算法计算形状上下文特征描述符的相似度。通过在MPEG-7, Kimia以及Swedish Leaf形状数据库上的实验结果表明,该算法对变形目标具有良好的鲁棒性,且提高了运算效率和匹配精度。此外,改进的离散曲线演化算法可与不同形状描述子融合形成新的形状匹配算法。%To solve the problem of contour noise and deformation in shape matching, a new shape matching algorithm combining N-DCE, SC and DP for retrieval accuracy and computational efficiency, is proposed based on discrete curve evolution(DCE). This algorithm includes 3 steps: the object contour evolved adap-tively by N-DCE and contour information including important visual parts extracted; the costing feature de-scriptor computed by Shape Contexts; shape distance similarity measured by DP algorithm based on SC costing descriptor. The experimental results of MPEG-7, Kimia and Swedish Leaf shape database indicate that this algorithm is robust to the contour deformation, and the computational efficiency and the retrieval accuracy are all essentially improved. It is worth mentioning that the modified discrete curve evolution al-gorithm can be fused with different shape descriptors to form a new shape matching algorithm.