The Research On The Calculation Of Topological Information And Discrimination Of Profiled Fiber

With the technology development and living standard improvement, profiled fibers have been found more and more applications in textile industry.The shape of profiled fiber affects the textile properties.In order to classify profiled fibers,shape feature extraction is needed to describe topological property.However,feature extraction methods proposed in literature are not suitable forprofiled fibers,a more effective method for profiled fiber's feature extractionis discussed in the paper.Skeleton is one of the most important methodsin representation and recognition of 2D objects.Skeleton-based representation contairs both shape features and topological structures of objects.Moreover,skeleton is easy saved as trees or graphs which are effective in object's shape matching.Some skeleton extraction algorithms have shortcomings. Thinning algorithms cause skeleton points shift from media-laxis and algorithms based on medial-axis transform can not guarantee the connectivity of the skeleton.These skeletons are sensitive to boundary deformation and many redundant branches are generated consequently. That may seriously interference the topology of the skeleton's graph. Therefore,such skeleton branches should be pruned.The pruning process will shorten all skeleton branches That cause local information missing and global feature error.Thus,a novel skeleton extraction approach is proposed in this paper.It is based upon vector inner-product,by which a well connected skeleton is obtained and no skeleton pruning is executed.Firstly,a sequence of border points is obtained by object contour tracking in binary image The distance from each point that enclosed by the sequence(called inner point)to the nearest border point is calculated by distance transform.The vector connectingthe two points is called the border vector.The inner-product is calculated between border vectors of inner point and its each eight neighboring point,and the smallest value of inner-product of each inner point is called local minimal inner-product. Secondly,a graph of local minimal inner-product is generated by calculating inner-product value of each inner point and the seeds of skeleton are selected by the value of inner-product and growing direction of seed points is determined according to the location of nearest border point corresponding to the skeleton seed point in bordersequence.The approach connecting two skeleton seed points is of smallest value of local minimal inner-product and is seek in minimal inner-product graphic with eight neighboring points search method.A well connected skeleton is determined by two steps of skeleton growth and it becomes single pixel wide after thinning algorithm.The proposed algorithm is proved tobe effective in skeleton extraction.The skeleton indicates accurate medial-axis position.It is stable for boundary noise.Also,the skeleton needs no pruning.With the information of skeleton endpoints and the branch points, the results of fiber classification show that these topological information parameters extracted from profiled fibers using theproposed algorithm are helpful in fiber recognition.Finally,the proposed algorithm is tested by MPEG-7 dataset.