Point Flow Edge Detection Method Based on Phase Congruency
Bing Bai, Fang Yang, Li Chai · 2019
This paper aims to detect image edges based on the Point Flow (PF) theory [14]. in the original PF model in [14], the vector field is built on image gradient. However, due to the noise sensitivity and weak edges confusion of local image gradients, the PF model might fail in detecting low-contrast edges. In this paper, we propose to build the vector field based on phase congruency (PC), which exploits image non-continuities (e.g., edges) in frequency domain instead of image domain. Since the phase of each frequency component at signal non-continuities is identical, e.g., after Fourier transform, the phase of all sinusoidal wave at rising edges of the square wave is 0circ, phase congruency indicates the image edges in frequency domain and avoids the effect of signal local contrast. Natural and synthetic images are tested to investigate the effectiveness of the proposed phase congruency based point flow model (PCPF). Compared with the classical Canny operator, the gpb-ucm detector and the original point flow model [14], our method yields higher detection accuracy and more detail edges.