相位一致性的理解及两种新的相位一致性模型  

Understanding of phase congruence and two new phase congruence models

作　　者：汪剑鸣[1,2] 窦汝振[3] 王中伟[1] 闫志杰[1] 

机构地区：[1]天津工业大学信息与通信工程学院,天津300160 [2]卡耐基梅隆大学统计系,匹兹堡宾夕法尼亚州美国15213 [3]中国汽车技术研究中心,天津300162

出　　处：《计算机应用研究》2010年第5期1948-1951,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(60602036)

摘　　要：首先研究了1D方波信号边缘处的相位分布特点,发现不同频率分量在边缘处的相位不一定收敛于同一个角度值,而是会随着占空比的变化而变化。因此,基于局部能量的相位一致性模型实际上是不同频率分量利用幅度加权后的平均结果。由于在多数自然图像中,图像频率分量的幅度会随着频率的增加而呈反比下降,基于局部能量的相位一致性模型主要由低频分量决定,这样在一定程度上降低了对边缘空间定位的精度。为了提高边缘检测的空间分辨力,提出了两种新的相位一致性模型。利用二维Log Gabor滤波器提取图像的局部相位信息,分析了利用这三种相位一致性模型进行边缘检测的结果。实验表明,利用两种新的相位一致性模型可以检测到更加细致的图像边缘信息,具有更高的边缘空间定位能力。This paper analyzed phase congruence at 1D step edges, where rectangular waves with different duty cycles were considered. It found that phase congruence at a step edge varies with duty cycle where all phases of different frequency components were not necessary to converge to a specific angle. Based on this fact, local energy based phase congruence could be regarded as an averaged result over different Fourier components weighted by their amplitude. Since the amplitude of a natural image often falls off at a rate that is roughly inversely proportional to frequency, the model mainly depends on low frequency components, which will lower the edge location precision. In order to get edge detectors of higher spatial precision, proposed two new phase congruence models. To testify the new models, adopted 2D Log Gabor filters to extract local phase, and applied all three phase congruence models to images for edge detection. The experimental results show the new phase congruence models have higher edge localization precision than that of local energy model and can tell us more edge details in images.

关 键 词：相位一致性 计算机视觉 图像处理 边缘检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]