On the choice of filters bank for phase congruency computing in noisy images

Phase congruency (PC) shows a noticeable sensitivity to noise. In this paper, an evaluation of phase congruency edge detector for very noisy images using three filters bank is presented. Firstly, we present two usual filters bank used to compute phase congruency which are Stretched Gabor filters (SGF) based on new angular component and Gaussian based quadrature pairs of filters (GQPF). The performance of both filters is compared to log Gabor filters (LGF) across a range of underlying noise levels. Thus, synthetic and real images corrupted with Gaussian noise are used in our experiments. The performance of the three filters to extract correct edge pixels is also carried out. Finally, simulation results are performed to illustrate the better choice of filters bank for phase congruency computing in noisy images.