Greedy Algorithm for Fast Finding Curvilinear Symmetry of Binary Raster Images

The article proposes a fast method for detecting curved symmetry for binary images by greedily searching for locally symmetric nodes of a polyline inside a figure, starting from a user-specified point. The advantage is that the procedure is virtually devoid of any manually adjusted inputs. The key procedure inputs (the increment and angular range used at the next step) are estimated by an adaptive procedure. Further, the obtained fragments of the image corresponding to the found segments of the polyline are sequentially superimposed on a single axis by rotation, forming a “straightened” figure, which makes it possible to calculate the Jaccard measure of symmetry for the entire figure. The experimental results demonstrate the successful operation of the method even on images with significant curvature at a perfect calculation speed.