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：92-96
出版社：International Journal of Soft Computing & Engineering
摘要：Scale-space theory provides a well-founded framework for modelling image structures at multiple scales, and the output from the scale-space representation can be used as input to a large variety of visual modules. Visual operations such as feature detection, feature classification, stereo matching, motion estimation, shape cues and image-based recognition can be expressed directly in terms of (possibly non-linear) combinations of Gaussian derivatives at multiple scales. In this sense, scale-space representation can serve as a basis for early vision. The Gaussian scale-space is widely used to model the human visual system. The main reason why Gaussian scale-space solely being used is that the Gaussian function is the unique kernels which satisfies the causality property i.e., it states that no new feature points are created as the scale increasing. The Gaussian filter are highly suitable for smoothing image. The amount of smoothing depends on the value of the standard deviation parameter of the Gaussian function. The problem of creating Gaussian scale- space is that if image smoothing does not stop in a proper point; it may lead to extreme destruction of local features of the image. In this paper, an approach has been presented to enhance a scale-space based on Gaussian function in order that a threshold is chosen for standard deviation in Gaussian filter with aim of preventing extreme destruction of image local features. Results of the proposed method indicate that this method affects considerably prevention of extreme destruction of image features and it can be very effective on creation of scale-space with high accuracy.
