Point Flow Edge Detection Method Based On Phase Congruency

Edge detection is one of the key technologies in image processing.It has a wide range of applications in the fields of biomedicine,medical facilities,industrial inspection,robot vision,and so on.Extracting accurate and complete edges is of great significance to the subsequent work of image processing.Existing edge detection methods have the disadvantages of poor noise resistance and inaccurate positioning.Aiming at these disadvantages,based on the point flow(PF)method based on gradient,this thesis proposes two new point flow methods based on phase congruency:The first method is point flow method based on single phase congruency(PCPF).We use single phase congruency to construct the new vector field of point flow model in the method.The second method is point flow method based on multi-scale phase congruency(MPCPF).We apply multi-scale phase congruency to contrust the vector field in this method.The specific work is as follows:1.The PCPF method is on the basis of the PF method.The image gradient is sensitive to noise.However,Phase congruency has rich image information,is not affected by image contrast and brightness,and has anti-noise performance.We use phase congruency to construct the new vector field.Then,the selected points flow under the action of the new vector field.Moreover,we add an edge integration step to removes small connected domains and short line segments after the point flow movement.Finally the trajectories is approximately equivalent the image edge.Compared with the Canny operator,the gPb?ucm operator,and the original PF method,our method guarantees noise immunity and robustness to low-contrast images.2.Because phase congruency has more texture information,the PCPF method is not effective for the detection of images with complex backgrounds.Multi-scale fusion may enhance strong edges while suppressing texture information,so the article proposes the MPCPF method to solve the problem.Firstly,phase congruency is multiscaled through the image pyramid.Then we use fused phase congruency to construct the vector field.Next,the selected points flow under the action of the new vector field.Finally,The image edge is obtained by the trajectories of the point motion.Experiments prove that the method has noise resistance and performs better on edge detection of images with complex backgrounds.