精确检测图像边缘特征是进行高分辨率遥感图像分割和识别的关键。空域特征检测算子以解决阶跃形边缘为主,得到的边缘特征对图像的亮度和对比度敏感。本文引入了一种基于频域相位一致的图像特征检测方法,该方法对遥感图像亮度和对比度具有不变性,同时适用于多种边缘特征的检测。使用Log Gabor小波计算IKONOS Pan图像的相位一致多尺度梯度,对农田、道路和厂房等典型地物进行特征检测的结果表明,相位一致算法对图像局部亮度和对比度不敏感;并且对线形物体产生单线响应,不似空域检测算子产生双线响应。最后考察滤波器尺度和方向参数变化及添加高斯噪声对检测结果的影响,发现相位一致算法无需先使用低通滤波去除噪声,因而具有更稳定的特征定位精度;并且抗噪声干扰的能力强,检测结果不会因为噪声而出现波动。基于相位一致的遥感图像不变特征提取,为高分辨率遥感图像的分割和对象识别提供了基础。

Phase Congruency was introduced as a new method to detect features from high-resolution remotely sensed imagery. Three objects i.e. paddy, road, and factory, were selected from the IKONOS Pan imagery of Nanjing. The Phase Congruency feature images were obtained by applying Phase Congruency model to the images with two-octave Log Gabor wavelets filters over 5 scales and 6 orientations. For qualitative comparison, the outputs of the Sobel and Canny detector were also presented. The results show that Phase Congruency response value was invariant in image local illumination and contrast. The other obvious difference was that the Sobel and Canny detector produced responses on each side of linear features, whereas the Phase Congruency produced a response centered on the line. The effects of varying filter orientation or scale and the effect of adding noise in the feature detection between Phase Congruency and Canny method then were compared. It was found that Phase Congruency could obtain more accurate localization because it does not need low-pass filter to restrain noise first. It was also shown that noise had been successfully ignored in the smooth regions of the image, unlike the Canny detector results fluctuated all over the image. With this invariant image feature detector, better image segmentation and object-oriented image recognition are possible.