基于Gabor滤波器组的多特征尺度不变特征提取方法

Invariant Feature with Multi-Characteristic Scales Using Gabor Filter Bank

提出了一种多特征尺度不变特征提取方法,简称GIFT(Gabor scale-Invariant Feature Transform).该方法首先利用2D Gabor滤波器组模拟生物视觉感知计算模型进行特征点检测,符合生物视觉感知特性,得到具物理直观性、稳健的特征点.其次采用基于Gabor核函数的特征尺度选择方法对所检测的特征点选择多个特征尺度,得到高可区分性的多特征描述子.最后,通过设计面向多特征尺度的特征匹配策略,提高特征匹配的可靠性.基于标准数据集的对比实验结果表明,GIFT方法在特征匹配率和稳健性上均优于SIFT.

An invariant feature extraction method called GIFT（Gabor scale-Invariant Feature Transform） is proposed,which has multi-characteristic scales.Firstly,by using 2D Gabor filter bank to model the biological cognitive computational model,intuitional and robust keypoints are detected.It is accordant with vision perceptive characteristic.Secondly,multi-characteristic scales of the detected keypoints are selected based on the Gabor kernel function,and then multi-characteristic descriptors with high distinctiveness are obtained.Finally,a feature matching strategy for multi-characteristic scales is designed,which increases the reliability of feature matching.The comparison experimental results on the standard dataset show that the proposed GIFT outperforms SIFT on both feature matching rate and robustness.