Chinese Journal of Computers   Full Text
Title	Edge Detection Based on Density Gradient
Authors	SUN Da LIU Jia-Feng TANG Xiang-Long
Address	(School of Computer Science and Technology, Harbin Institute of Technology, Harbin 150001)
Year	2009
Issue	No.2(299—307)
Abstract &   Background	Abstract This paper proposes a new image edge detection algorithm based on the density gradient. The image pixels are taken as some sample points in the image sample space, and the image can be transformed into the density gradient field by density gradient estimation. In the density gradient field, all the density gradients point against the image edge. According this property, the new algorithm selects the points around which all the gradients point against them as the edge points. Comparing with other detectors, the edges detected by the new algorithm are invariant to the detection scales, and the object’s boundary can be directly selected out by the new detector as their size. Keywords edge detection; density gradient; image processing Background Edge is one of the basic features to describe the geometrical information of objects, and edge detection is a very important task for image analysis and image understanding. Many edge detection methods have been proposed, but there is no an edge detection method suitable for all type of images yet. And in the edge map detected by these methods, the corners of the objects are badly distorted with large detection scales. The density gradient based method proposed in this paper is applicable to all type of images: gray level, color and multispectral image. And the novel method can preserve the corners especially in large scale. The new method belongs to the project “The research of human motion tracking with the motion camera by the MCMC particle filters”, which is supported by the National Natural Science Foundation of P.R. China (60672090). Although many tracking method is developed, there is only a few researches about the rotation camera tracking. The new rotation camera tracking system will be developed in this project. The precise edges detected by the density gradient based method will be used to matching and recognizing human body. The new method can precisely find the large object’s boundary directly.