摘要
针对传统局部特征提取算法在提取特征点时效率不高,生成描述子需要计算主方向等问题,结合SURF算法和RGT(Radial Gradient Transform),在精度损失尽可能小的情况下提高局部不变特征提取速度,提出一种改进的AR-SURF(加速径向SURF)算法。该方法在特征检测阶段,在定位特征点时减少构造尺度空间时所计算的响应层个数,将求取对应点响应放在定位阶段。在特征描述阶段,取消确定主方向的过程,将特征点周围区域的Haar小波响应进行RGT变换,然后将特征点周围区域划分为多个同心圆,并统计特征点周围圆形区域内的响应结果,最后利用小波响应结果得到旋转不变的特征描述子。实验结果表明,AR-SURF算法节省了时空损耗,提升了定位速度,提取效果更好,更加合适于海量图片处理。
For some disadvantages existing in traditional local feature extraction algorithm,such as the inefficiency when extracting feature points and the need to calculate principal direction when generating descriptor,etc.,we propose in this paper an improved accelerated radial SURF algorithm by combining SURF algorithm and RGT（ radial gradient transform） and speeding up the process of local invariant feature extraction in the circumstance of lesser precision loss as much as possible. In the step of feature detection,the algorithm decreases the number of response layers calculated in constructing the dimension space when locating the feature points,and places the course of corresponding point calculation in localisation phase. In the step of feature description,it cancels the process of determining principal direction,and conducts RGT transformation on Haar wavelet response in surrounding regions of feature point,then divides these regions into concentric circles and counts the response results within the surrounding circle regions of feature points as well,finally it uses these wavelet responding results to obtain the rotation-invariant feature descriptors. Experimental result demonstrates that AR-SURF algorithm saves the loss of time and space,increases the speed of localisation with better extraction effect,so it is more suitable for mass images processing.
出处
《计算机应用与软件》
CSCD
2016年第4期187-191,共5页
Computer Applications and Software
基金
国家自然科学基金项目(61462069)
内蒙古自然科学基金项目(2014MS0622)
内蒙古科技大学校内基金项目(2011NCL054)
