%0 Journal Article
%T 基于非参数边缘积分估计的局部特征描述子<br>Local feature descriptor based on nonparametric marginal integration estimation
%A 郑明国
%A 吴成东
%A 陈东岳
%A 贾同
%A 金勇虎
%J 北京航空航天大学学报
%D 2016
%R 10.13700/j.bh.1001-5965.2015.0156
%X 摘要 为提高图像匹配性能提出了关于局部区域特征描述子的统计模型。该模型是一种基于梯度模值及方向分布的边缘积分函数模型。在离散梯度方向的边缘积分函数与梯度矢量场的模值累积方向直方图相同。采用基于核函数的非参数估计,估计了该函数,应用于尺度不变特征变换(SIFT)描述子。为了提高描述子的旋转不变性、独特性,降低运算复杂度,将特征点周围的局部区域作为圆形,由径向采样网格划分为8个子区域。在每个子区域估计边缘积分函数,特征向量由每个小块8个方向的函数值组成。实验表明,该描述子能够提高旋转变换的检测率(查全率),降低运算复杂度。<br>Abstract：A statistical model for the feature descriptor of local region was suggested to improve the image matching performance. This model is a marginal integration function model based on the gradient magnitude and orientation distribution. The marginal integration function on the discrete gradient orientations is the same as the magnitude accumulation orientation histogram of gradient vector field. Using the nonparametric estimator based on kernel function, we estimated this function and applied it to scale invariant feature transform (SIFT) descriptor. To enhance rotation invariance and distinctiveness and to reduce computational complexity for descriptor, local region around the feature point was selected as circle and partitioned to the 8 sub-regions by radial sampling grid. The marginal expectation functions are estimated in each sub-region and the feature vector consists of the function values on the 8 orientations for 8 sub-regions. Experiments show that this descriptor can improve detective rate (recall) for rotation and reduce computational complexity.
%K 梯度分布
%K 非参数估计
%K 边缘积分
%K 尺度不变特征变换 (SIFT)
%K 图像匹配&lt
%K br&gt
%K gradient distribution
%K nonparametric estimation
%K marginal integration
%K scale invariant feature transform (SIFT)
%K image matching
%U http://bhxb.buaa.edu.cn/CN/abstract/abstract13471.shtml