一种新的结合环状区域划分的特征描述子

Local coordinate gradient descriptor based on cyclic division

为提高图像特征描述子的仿射不变性及匹配性能,提出了一种特征区域描述方法.首先,该方法用Hessian-affine特征检测子提取特征区域,将其归一化为单位圆特征区域以满足仿射不变性;然后,采用环形区域划分方法且在图像样本点的局部坐标系上统计梯度方向,这样,不仅满足了旋转不变性,而且避免了直方图统计主方向带来的误差.实验结果表明,新方法不仅具有良好的仿射不变性,而且在匹配性能上明显优于尺度不变特征描述子.

A novel regional image feature description method is proposed to improve the affine invariant and matching performance of image feature descriptors.Firstly,the Hessian-affine detector is adopted to extract the image feature region,which is normalized to the unit circular feature region to meet the affine invariant requirement.Secondly,in order to avoid the error caused by calculating main gradient orientation on the histogram,the annular-based region division is proposed on the local coordinate system of the image sample point to calculate the local gradient orientation.Experimental results demonstrate that the new method not only has an excellent affine invariant,but also is significantly better than the SIFT descriptor in match performance.