The algorithm of image features detection from phase congruency model based on 2-D Hilbert transform

The algorithm of phase congruency developed from phase information of image in frequency domain, is employed for image feature detection. The theory of phase congruency is that image features, such as step edge, roof, Mach band and delta, always occur at points where the phases of harmonic components come to the maximum congruency. This algorithm is realized to extract the image features by constructing the local energy model being normalized to get the value of phase congruency of every point in image. This paper introduces the 2-D Hilbert transform instead of 1-D Hilbert transform to propose the algorithm of phase congruency for detecting the image features. The modified algorithm can take account of the full directions of the image features. Meanwhile, the proposed method simplifies the calculation of the numerator of local energy by convoluting the original image with the window operator to remove DC (Direct Current) component of current window and 2-D Hilbert transform respectively. Moreover, this algorithm makes the denominator of the model of phase congruency adding the DC component to suppress the noise of image. The modified algorithm of phase congruency is implemented into the natural image and remotely sensed imagery, and the results show that the modified algorithm is effective to detect image features.