Extracting closed object contour in the image: remove, connect and fit

Abstract

Contour extraction is one of the fundamental problems in computer vision. How to extract closed object contours in noisy images is an interesting challenge, which is not solved well by current methods. In this paper, a method of extracting closed object contours through removing, connecting and fitting is proposed. Firstly, existing preprocessing steps are employed to produce a set of contour segments from an image. Secondly, an 8-neighborhoods discriminant is advised, which is used to determine and remove the nontarget curve pieces. Thirdly, a connection algorithm based on proximity and continuity of closed contours is presented to connect the fractured curve segments to form a closed object contour. Fourthly, a B-spline curve-fitting method is provided to make the closed object contour more consistent to the object’s real contour. Finally, real applications and comparative experiments are conducted to testify the proposed method’s performance, effectiveness and robustness. The comparison shows that the proposed method can obtain a better closed contour even in a noisy image.