ALGORITHMS FOR ADJUSTMENT OF SYMMETRY AXIS FOUND FOR 2D SHAPES BY THE SKELETON COMPARISON METHOD
Reflection symmetry detection for 2D shapes is a well-known task in Computer Vision, but there is a limited number of efficient and effective methods for its solution. Our previously proposed approach based on pair-wise comparison of sub-sequences of skeleton primitives finds the axis of symmetry within few seconds. In order to evaluate the value of symmetry relative to the found axis we use the Jaccard similarity measure. It is applied to the pixels subsets of a shape which are split by the axis. Often an axis found by the skeleton comparison method diverges more or less from the ground-truth axis found by the method of exhaustive search among all the potential candidates. That is why the algorithms that allow adjusting the axis found by the fast skeleton method are proposed. They are based on the idea of searching the axis which is located near the seed skeleton axis and has greater Jaccard similarity measure. The experimental study on the ”Flavia” and ”Butterflies” datasets shows that proposed algorithms find the ground-truth axis (or the axis which has slightly less Jaccard similarity value than the ground-truth axis) in near real time. It is considerably faster than any of the optimized brute-force methods.