【作者】 张倩；

【导师】 程建；

【作者基本信息】 电子科技大学 ， 工程硕士（专业学位）， 2017， 硕士

【摘要】 图像的有效表示是图像处理任务中的主要问题之一。Smoothlet作为一种自适应多尺度和几何表示方法,可以自适应地表达不同的位置、尺度、方向、曲率和边缘平滑的图像。Smoothlet变换在使用过渡带定义模糊的边缘时,在水平或竖直方向平移边缘曲线得到的区域即为过渡带。尽管Smoothlet变换可以对边缘进行曲线形表达,并且可以表示模糊的边缘,但是由于现实图像中的灰度图像的边缘可能很复杂,仅仅在水平或竖直方向上定义过渡带不足够表达真实图像边缘,因此需要对Smoothlet进行改进。本文对基于Smoothlet的变换进行改进,主要研究内容如下:1.提出Smoothlet变换的改进框架——Extended Smoothlet(Ex Smoothlet)。该框架将曲线平移的方向定义为合理的方向,并提出了该框架的两种实现算法。提出了Elliptical Ex Smoothlet(EES)算法,与Smoothlet相比,过渡带所有点的平移方向统一采用直线参数的法线方向。基于Fast Wedgelet实现了EES的变换——Fast Elliptical ExSmoothlet Transform(FEEST),将其与传统Smoothlet变换的快速算法Fast Smoothlet Transform(FST)相比,实验结果表明,FEEST比FST对图像近似和去噪的效果都要更好。2.提出Homocentric Elliptical ExSmoothlet(HEES)。当EES与Smoothlet使用椭圆模型作为边缘曲线时,为了使得两条边缘曲线不相交,只能使用半椭圆(椭圆上部分或下部分)的其中一部分,当真实图像边缘延续超过半椭圆时,EES与Smoothlet都不能准确表示。HEES使用椭圆拟合方法可以拟合椭圆的任何部分,并在过渡带中沿着椭圆中心方向平移边缘曲线,能对任意延续椭圆边缘进行描述。在HEES的应用方面,将其与EES相比,实验结果表明,HEES比EES对图像近似和边缘检测的效果都要更好,对噪声小的图像效果也好。更多还原

【Abstract】 An effective representation is one of the main issues in image processing tasks.Smoothlet is an adaptive multi-scale and geometric representation method,which can adaptively expresses different positions,scales,directions,curvature and smooth edges of image.Smoothlet transform uses extruded surface to define a blurred edge,which can be obtained by translating the edge curve in the horizontal or vertical direction.Although the Smoothlet transform can represent various blurred edge,the edge of the grayscale image in the real image may be complex.And the extruded surface that is only defined in the horizontal or vertical direction is not enough to express the real image edge.So it’s necessary to improve the Smoothlet.In this paper,new transform is used to improve the Smoothlet transform.The main contents are as follows:1.An improved framework for Smoothlet transform-Extended Smoothlet(ExSmoothlet)is proposed.The framework defines the direction of the curve as a reasonable direction.And two implementations of the framework are proposed.The Elliptical ExSmoothlet(EES)algorithm is proposed in which the translation direction of all points in the transition zone is unified by the normal direction of the straight line parameters.Based on the fast Wedgelet,the Fast Elliptical ExSmoothlet Transform(FEEST)is implemented based on the Fast Wedgelet.FEEST is compared with the Fast Smoothlet Transform(FST)algorithm.Experimental results show that FEEST is better than FST for image approximation and denoising in PSNR sense.2.Homocentric Elliptical ExSmoothlet(HEES)is presented.When EES and Smoothlet use the elliptical model for the edge curve,only a part of the semi-ellipse(part or lower part of the ellipse)can be used in order to make the two edge curves do not intersect.When the real image edge continues beyond the semi-ellipse,the EES and Smoothlet can’t be accurately expressed.HEES use an elliptic fitting method to fit any part of an ellipse and define trasition direction in the adaptive way.In the application of HEES,the experimental results show that HEES is better than EES for image approximation,edge detection and image denoising with lower noise.更多还原