A Study Of Distance Transform Based Fiber Skeleton Determination Algorithm

Shape representation and description plays an important role in most computer vision research. Skeleton displays crucial information about the shape and has many applications in pattern recognition, shape analysis and object identification. In the research proposed here, the skeleton calculation is implemented to realize the fiber classification. Edge nodes are detached and the connected skeleton is obtained by the thinning algorithm under the condition that the region connection is preserved. The thinning algorithm is an iterative calculation that needs heavy computation. Moreover, the precision of skeleton is not proved to be satisfactory.Distance transformation is a fundamental computation used to determine the distance map for an input image. In order to overcome the shortcomings of thinning algorithm, a distance transform based skeleton algorithm is proposed to determine the profiled fiber's skeleton with the minimal cover set. Firstly, the distance transformation is executed for the binary image. Secondly, the set of local center points is decided according to the relation with the neighboring elements. The edge information of binary image is applied to build a correlated matrix together with the local center point set. Minimal cover set is determined with the transform along the row and column directions. A connecting process is used because the medial points are generally isolated. An uphill algorithm starts from such points that have the least distance value, and advances along the direction of the steepest ascending for each medial point and saddle point in the distance map. The calculation stops when another skeleton point is encountered. The skeleton is sensitive to boundary noise; therefore, process should be applied to suppress such interference. In order to improve the performance of the profiled fiber recognition, pruning regularization should be used. The experimental results demonstrate that the algorithm proposed in the paper is satisfactory for persevering of the topology information.The skeleton algorithm based on distance transformation is applied to the classification of profiled fiber. The recognition ration is similar to thinning algorithm by the test upon the bell, hollow and W fibers. But it is improved apparently by the test on the trilobal, cross and star fibers. It is shown that the proposed algorithm is appropriate for the profiled fiber with a 95.4% recognition ration.