A Modified Thinning Framework Against Noise

Abstract

We proposed a modified thinning framework that based on the scale space technique to automatically extract skeletons from images without manual-tuning. The proposed framework can increase the robustness of the thinning algorithm, it not only can suppress the boundary noise, but also can alleviate the inner noise. These two types of noise generally cause the appearance of the abundant of the unwanted branches in the outcome of the thinning algorithm, which arise the difficulties of the later recognition or matching process in skeleton. The experiment has proved the proposed framework has better performance when comparing with the other existing methods.