Tracing on heterogeneous grids to improve the concavity performance of snake algorithms

Existing contour tracing algorithms operate on binary images at pixel level to traverse the boundary of an object. However, if the original grayscale image has obscure edge transitions, it can be very challenging to extract binary objects properly for pixelwise tracing. A more reliable approach can be to consider a rough approximation of the objects and perform tracing on that, followed by a final adjustment. We can use e.g. the quadtree representation which belongs to the family of heterogeneous grid representations. In this paper, we show how a contour tracing algorithm operating at pixel level can be extended to a heterogeneous grid representation. The robustness of our method is demonstrated in snake algorithms to improve their concavity performance.