一种旋转不变特征描述符

A rotation invariant feature descriptor

提出一种基于图像梯度旋转直方图(RHG,rotation histogram of gradients)的快速计算旋转不变特征描述符算法。RHG描述符使用直方图旋转的方法获得旋转不变性,采用直方图加权合并的方法降低边界效应引起的描述符统计矢量的突变。RHG描述符将特征点主方向的计算与描述符的计算合并,提高了计算效率。RHG描述符在图像存在尺度改变、3维视角变化引起的变形、旋转变化、照度改变和噪声等因素的影响下,具有较强的鲁棒性。RHG描述符的性能与尺度不变特征变换(SIFT,scale invariant feature transform)描述符相近,但计算速度提高2倍以上。

A fast rotation histogram of gradients（RHG） feature descriptor is presented.Usually,to achieve rotation invariance,the main orientation of a feature descriptor is figured out,and then the coordinates of the sample region and the gradient orientations are rotated around the main orientation with low efficiency.The RHG descriptor can achieve rotation invariance by rotating the bins of its gradient histogram and avoid boundary effect by the combination of bins.The step of main orientation is also merged into the RHG descriptor for efficiency.RHG descriptor is invariant to image scale and rotation,and it can provide robust matching across a substantial range of affine distortion,change in 3D viewpoint,addition of noise,and change in illumination.