The Research Of The Skeleton Extraction From Local Image Based On Level Set Theory

The extraction of image features is a preliminary step for image analysis. As a linear feature of the image, skeleton can simplify the representation of the shape and has important applications in image matching and recognition. The stability is usually required for a good algorithm of skeleton extraction because the skeleton depends on the edge and is easily affected by the shape perturbation. In addition, it’s a challenge to determine the position of the object with the interested skeleton in the image due to the structural complexity of the image.In recent years researchers have mainly focused on the study of the skeleton extraction method in the graphics and less in the gray image. The existing methods in gray images mainly aim to obtain the skeleton of ribbon objects directly, while the purpose of this thesis is to study the robust algorithm which may extract the skeleton of a specified object with general shape. In this paper, a two-step algorithm, segmenting the specified object to ascertain its position firstly and extracting the skeleton secondly, is provided based on the previous researches. It is proposed by combining the level set method with the image segmentation and the geometric features of the skeleton. The details of this paper are listed as follows:Firstly, an energy function is constructed in the piece-wise constant image through analyzing the change of the gray values distribution in the segmented domain during the contour evolving. Then it is converted to the level set equation by applying the variation method. Moreover, the model is improved with a step-based approach and extended to the general gray image by analyzing factors hindering the contour evolving. Only the object interested is singled out after applying the model to the gray image, and it also shows robustness for the insensitivity to the noise in the image.Secondly, the signed distance function of the segmented edge belongs to a special level set equation, Eikonal equation, and it can be solved by the fast sweeping method. There is a close relationship between the skeleton and the signed distance function, so a coarse skeleton is obtained by using the property that grads modules of signed distance function do not equal to one at skeleton points. And then we trace the endpoints of the skeleton along the normal direction on the corner points after studying the corner detection method and the geometric characteristic of the endpoints. Finally the skeleton is obtained by utilizing the shortest path algorithm to connect the endpoints. The algorithm is invariant to the image rotation and the edge perturbation, which will make it indifferent to the slight error of the result in the segmentation step.At last, the two-step algorithm is respectively applied to the gray images with and without noises. Simulation experiments demonstrated that smooth skeletons with one pixel width can be obtained by our method. And the results have high similarities of the shape and position, which verifies the advantage of the algorithm again. Besides, it provides a solid foundation for the matching and recognition of the specified target in the gray image since we only extract the skeleton of the interested object.