Aiming at the shortcomings of traditional corners,a novel junction point detection algorithm which can extract local and structural information is proposed.First,sub-pixel positions of edge points are obtained using non-maximal suppression operation carried out in the gradient magnitude image according to gradient directions.Second,noises and irrelevant edges are filtered out based on azimuth consensus,and the remaining edge points are grouped relied on their inclination angles.And then,sup-pixel location of a junction points is found by a refinement strategy applied to the initial integer location of the junction.Finally,the inclination angles of edges points are updated according to the refined location of the junction,and optimal inclination angles is computed.Experimental results show that the proposed algorithm has good accuracy both in positioning and in inclination angles of branch edges,and has satisfactory robustness to noise disturbance and contrast change.%针对传统角点的缺陷,提出一种能够提取局部结构信息的分支点检测算法.首先在梯度强度图像中按照梯度方向进行非最大抑制,得到边缘点的亚像素位置；然后利用方位一致性滤除噪声及不相关边缘的干扰,并根据倾角方向对剩余的边缘点进行分组；再对分支点的整数位置进行求精,得到分支点的亚像素位置；最后根据分支点的最优位置更新边缘点倾角,计算最优的分支边缘倾角.实验结果证明,该算法具有较好的分支点定位精度和分支边缘倾角精度,对噪声干扰及对比度变化具有较好的鲁棒性.
展开▼
