Contour abstraction based on salient points

Contour representation is an important application in image compression, template matching, object detection and recognition. However, it is far from meeting the current requirement due to the expensive computational cost and complex noise in the real-world application. In order to make contour representation more practical, we propose a novel approach of abstracting contours of the objects in an image. In our approach, we firstly find the salient points on the target contour by combining an ellipse model and Chord-to-point distance accumulation techniques. Then, based on the salient points, we adopt the least square method to fit a planar arc representing the target contour. The extensive experiments show that our approach has the lower computation cost, better robustness and more exact approximation to the original target contour. Our work provides more selections for the practical application of contour abstraction.