A Phase Congruency Based Corner Detector for Images under Different Illuminations

Reliable corner detection is extremely important for applications such as 3D scene reconstruction, image registration and motion tracking. There are many kinds of corner detector which have been wildly used. However, the image´s changing contrast and brightness under different illuminations may considerably influence these traditional detectors´ response, which makes the setting of universal threshold very difficult. In this paper, a phase congruency based corner detection method is proposed. The new operator finds the corner points within an image through searching the maximal phase congruency in frequency domain. The phase information is obtained by wavelet transform with Log-Gabor filter. The result of detection is highly localized and invariant to image´s illumination. Contrast experiments between the proposed detector and the widely used Harris detector have been made on both synthetic and natural scene images. The experimental results have shown that accuracy of detection and localization of corner points can be achieved with the proposed method.