Image Skeleton Extraction Based On Improved K-segments Principal Curves Algorithm

Image recognition is a technology, which can do image processing, image analysis and image understanding by use of computer to identify targets and objects of different models. Image recognition is based on different features of the image, such as edge, line, shape, color and texture, etc. As a basic feature of image, shape has unique advantages in the physical description and recognition. The skeleton is a simplified description of shape, which will greatly reduce the workload in the process of image recognition. And the skeleton is a kind of compression of the original graphics. It has same topological invariance in translation, rotation and scale, which can be useful for image recognition. So skeleton can be the research emphasis of computer vision, artificial intelligence, image processing and so on.In this thesis, we analyze and study the existing skeleton extraction algorithms, and propose a method of extracting skeleton of the hand drawn characters based on K-segments principal curves. K-segments algorithm can exhibit good performance when data are concentrated around a highly curved or self-intersecting curve, and truly reflect the form of data and maintain the information of it perfectly. The hand drawn character images are arbitrary and personalized, so it is difficult to identify. The paper chooses this kind of image as the experimental image to extract the skeleton. With these characteristics of principal curves, firstly, giving full consideration to the number of pixels and the degree of bending in the hand drawn character image, this method sets up optimal parameters to construct the corresponding principal components. Secondly, with the Hamiltonian path principle, it connects the principal components and optimizes them to Polygonal lines. In the end, the penalty term is used to implement a preference for ’smooth’ (not having sharp turns) curves. But there has some defects of this method, and some improvements have been made. After removing the redundant false edges, a better improvement is made that the method connect positions where are supposed to be connected rather than separated. Finally, the skeletons we get are used for topological similarity measurement, and the correctness of the proposed algorithm is verified.In this paper, K-segments principal curve is applied to image edge detection. Firstly, the image is processed to get the fuzzy edge region. Then the contour is extracted by the K-segments algorithm. And the skeleton of the fuzzy region can be considered as the boundary of the original image.