基于多尺度Gabor滤波器组的不变特征点提取新方法

 针对复杂背景的光学遥感图像数据,提出一种基于尺度空间理论和多尺度Gabor滤波器组的目标局部不变特征提取方法.该方法优点有二:一是利用Gabor滤波器的带通特性和多通道特性可从目标图像中提取更符合视觉特性、更具物理直观性的特征点,增强了不变特征提取方法在照度变化、噪声和背景干扰下的稳健性;二是基于尺度空间理论设计的多尺度Gabor滤波器组,可使提取的特征点具有尺度不变性,增强了对遥感图像复杂几何畸变的适应性.

A New Invariant Feature Detector Based on Multi-scale Gabor Filter Bank

For interpretation of remote sensing optical images in clutter background,a local invariant feature detector, which combines multi-scale Gabor filter with scale-space theory, is proposed. Owing to the band-pass and multi-channel characteristic of Gabor filter, the invariant feature points extracted by our proposed approach are more intuitional and more accordant with vision perceptive model.Hence the invariant features perform more robust under the illumination change and the disturbance of noise and background. On the other hand,as the multi-scale Gabor filter banks are designed based on the scale-space theory,the feature points are scale invariant, which leads to the feature detector more adaptable to imaging geometric distortion.