中文摘要：

当前局部不变特征描述符主要针对刚体形变的图像匹配问题,但非刚体形变图像匹配的需求普遍存在且应用日趋广泛.非刚体形变的复杂性会导致同名特征点的局部支撑区域难以保持结构特性的相似性.构造更具针对性的局部不变特征描述符成为解决非刚体形变图像匹配问题的关键.针对如何准确地确定局部特征的主方向并划分局部支撑区域,提出了一种有向非刚体形变局部不变特征描述符及其构建方法.首先根据特征点的二阶矩阵估计其椭圆邻域并指示主方向,然后对局部支撑区域进行同性化处理,以消除各向异性形变的差异,最后把局部支撑区域加权嵌入到三维空间,用热核信号的形变不变性进行特征点局部支撑区域描述.对比实验结果表明,在非刚性形变和光度变化情况下,本文描述符的匹配正确率高于SIFT（ScaleInvariant Feature Transform）和GIH（Geodesic Intensity Histogram）描述符,且保持了较好的旋转不变性.

英文摘要：

Recently,most local invariant features for image matching problems center around the invariance of rigid deformation.However,there is a widespread and growing need of the image matching for the non-rigid deformations.The complexity of non-rigid deformation makes the corresponding points＇ local support area difficult to maintain the structure similarity.So how to build the more pertinent local invariant feature descriptor is the key to solving non-rigid deformation image matching problems.We,therefore,propose an oriented non-rigid deformation local invariant feature descriptor and its structuring method to better specialize the direction of the local feature and divide local support area.We first estimate the elliptical neighborhood of each feature point and indicate the main direction according to its second moment matrix,and then normalized the local support area to eliminate differences in the anisotropic deformation.Finally,we embed the local area into 3D space and construct a descriptor in terms of a heat kernel signature,which is invariant to deformation.Comparative experiments show that,under the non-rigid deformation and photometric changes,our algorithm maintains good rotation invariance and a higher matching accuracy compared with the SIFT（Scale Invariant Feature Transform） and GM（Geodesic Intensity Histogram） algorithm.