Fully parallel thinning with tolerance to boundary noise

摘要

A new fully parallel thinning algorithm is developed and evaluated in this paper to solve the noise spurs problem and preserve geometric properties efficiently. The algorithm not only prevents the excessive erosions but also lessens the creation of spurious end points for an image with boundary noise. When two input images are similar in shape but with boundary noise, our skeletons produced appear more consistent in topology as compared to those using other algorithms. Although a few additional neighbors other than 3 × 3 are considered in the deletability conditions, the smoothing procedure prior to thinning is avoided. The parallel thinning algorithm runs very fast and can be implemented in real time. Several English and Chinese characters and the difficult patterns often illustrated in the literature are also experimented to show the efficiency and consistency of our algorithm.