摘要：

针对图像识别、检索、配准中的特征提取与特征描述问题，利用加权图的邻接矩阵特征值分解方法，构造具有平移、缩放、旋转不变性的图像边界形状和空间关系描述子。理论分析与仿真实验结果表明，该算法在缩放和旋转条件下具有较强的鲁棒性和较好的类别可分离性，相比传统算法可以更好地进行图像描述，且计算量有所减小。

关键词: 边界形状, 空间关系, 特征值分解, 图像识别, 图像检索, 图像配准

Abstract:

Aiming at the problems of characteristic extraction and description in image recognition, retrieval and registration, by using eigenvalue decomposition approach for weighted graph, this paper constructs boundary shape descriptor and spatial relation descriptor based on metric matrix eigenvalue, which is invariant under translation, scaling and rotation. Analysis and experiments prove that the spatial relation descriptor is robust under translation, scaling, rotation and small noise, and it has good similarity discriminating ability. Compared with traditional algorithms, it can describe image better with less computation.

Key words: boundary shape, spatial relation, eigenvalue decomposition, image recognition, image retrieval, image registration

中图分类号: 

• TP391