Research On Salient Skeleton Computation Method For Image Objects

Skeleton is one of the simplest but powerful tools for shape representation and analysis.It encodes the original shape by one less dimension,but it can preserve the topological information well,which is convenient for computer to save and transfer the structural features of image.Therefore,it is widely used in various fields of computer vision and computer graphics,such as shape retrieval,object registration,human posture recognition,medical image processing and so on.Since the concept of skeleton was proposed,many excellent methods have emerged.Although there are many methods to get a better skeleton,there are also many disadvantages,such as being too sensitive to boundary noise,not ensuring the integrity of the topology,or requiring cumbersome post-processing operations.In addition,most of these methods are based on the premise of obtaining the complete boundary of the object,which is very harsh for the color image.To solve the above problems,we propose a new concept of salient skeleton.By analyzing the symmetry of image objects,we measure the probability that each pixel in the target object belongs to the skeleton point,generate the skeletal saliency map(salient skeleton),and then obtain the binary skeleton by the threshold.This method mainly has the following innovations:(1)we propose a new method of symmetry measurement,projection symmetric measurement,which fully considers the symmetry and global characteristics of the image object;(2)we propose a novel salient skeleton(skeletal saliency map)to encode the skeleton and provide the theoretical analysis of properties.(3)we give an efficient and robust implementation for binary images and color images,which directly reports a stable and noise-insensitive skeleton,rather than depend on the post-processing.In the implementation process,the method mainly includes the following steps:first,extract the boundary or probability edge of the target object in the image;then,for each pixel p in the target object,we project its four neighborhood pixels to the boundary,and then calculate the inferior arc distance between two projection points along the boundary,and select the maximum distance value of these inferior arcs as the projection symmetric measurement of the pixel p;after calculating the projection symmetric measurement of all the pixels in the target object,a skeletal saliency map is obtained,and then a binary skeleton is obtained by the binary operation.Our method can generate a two-pixel-width binary skeleton with the correct topological structure and avoid extra post-processing.Besides,the algorithm is robust to boundary noise.Extensive experiment results validate these nice properties of the proposed method,which shows that it outperforms the state-of-the-art methods.In addition,we also explore the application prospect of our method,and find that it can be applied to vector graphics generation,probability edge thinning and so on.