期刊名称：International Journal of Soft Computing & Engineering
电子版ISSN：2231-2307
出版年度：2014
卷号：4
期号：4
页码：1-7
出版社：International Journal of Soft Computing & Engineering
摘要：One of the basic requirements in images representation was the feature extraction and its proper description and has many applications in the image processing and the machine vision. Many of the local feature descriptors of image use the difference of Gaussian feature detector. This detector is too much invariant against the scale changes. In this paper, a procedure is presented to select a proper threshold for the standard deviation in Gaussian filter to improve the performance of difference of Gaussian detector. In this paper's method, based on the properties of co-occurrence matrixes, the spatial dependences between available points in the image are divided into three general classes: sharp points, middle points and unsharp points, and then, on the basis of this division, the appropriate position is determined for stopping the development of standard deviation in Gaussian filter in some way that it is prevented to destroy the sharp points in the image and also to select the noise points as the key points of image.
关键词：Difference of Gaussian (DOG); Feature;Detector; Interest Point; Key Point
