The Research Of Image Segmentation Method Based On K Principal Curves

Nowadays, with the appearance of information freeway and digitalearth, and with the wide application of Internet, the image information becomes the important resources and methods for getting and using information. Image segmentation is the important and foundational problem in the field of computer vision and image processing. There are many methods of image segmentation, but these could not meet the different demands of today. So the technology of image segmentation must to be developed and improved more. On the base of studying the classical methods, the paper proposes some new ideas.First, a new method based on DIS edge detection and adaptive edge growing technique for performing image segmentation tasks was presented. The method applies a difference in strength (DIS) technique to detect edges, DIS can obtain more complete edge information than gradient operator and Sobel operator etc. To solve the problem of having the wrong and discontinuous points, adaptive edge growing technique was proposed. After having connected the discontinuous points, the new adaptive edge growing produces some closed outlines. Finally, we attain the result of segmentation after region splitting and region merging which can avoid the phenomena of leakage and get rid of the fake edge. This method is an organic combo of edge detection, edge growing and region merging.The second method we proposed is based on AEP and the K principal curves. Principal curves are nonlinear generalizations of the first linear principal component. They emphasize for finding 'self-consistent' smooth nonparametric curves that pass through the middle of a multidimensional data set, and we can get the frameworkof the data set. The contours of objects are some variform curves which are closed, principal curves can realistically depict the shapes. First, the paper adopts the advanced erosion-propagation algorithm (AEP) combines the erosion-propagation algorithm and the region-growing version of watershed algorithm to classify the original image and label, produce the initial information of edge. Then according to the information of edge, the method links the edges of the objects using the polygonal line algorithm of K principal curves. In the end, we present the corresponding testing results of the two methods. The experiment shows both methods can provide accurately contours effectual.