【作者】 储昭辉；

【导师】 汪荣贵；

【作者基本信息】 合肥工业大学 ， 计算机应用技术， 2010， 硕士

【摘要】 图像增强作为图像处理中重要的预处理手段,由于其应用广泛,受到人们的广泛重视。所谓图像增强,是指有目的地强调图像像素的整体或局部取值特性,以改善图像视觉效果或满足特殊分析的需要。Retinex理论及相关算法是图像增强的新思路,其理论基础是色彩的恒常性,它通过模拟人眼观察场景的方式,恢复因图像采集设备限制而丢失的场景细节,达到增强图像对比度,还原物体真实色彩的目的。本文以Retinex算法为研究重点,通过对传统的Retinex算法的分析,利用小波变换多分辨率的特点,提出了一种新的Retinex算法,并将Retinex理论引入JPEG2000压缩框架,用以增强JPEG2000压缩图像。文章主要内容如下：1、通过从像素比较路径选择的角度出发,分析了Retinex算法一些常用的计算策略,包括基于随机路径的Retinex算法、基于迭代计算的McCann’s Retinex算法、基于中心环绕的Retinex算法,并进行了实验对比分析。然后对经典Retinex计算方法进行了分析,指出Retinex算法的特点和局限性,并在实验中得到验证。2、针对传统图像增强方法在图像增强的同时不能有效抑制噪声这一问题,本文根据小波变换多分辨率的特点,在Retinex理论基础上提出了图像亮度与噪声双估计模型的图像增强的新方法。该算法将图像小波分解后的低频系数进行波动抑制,对高频系数通过多子带独立系数模型进行亮度估计,同时使用不同的软-硬估计法对图像的噪声进行抑制。该方法不但能有效的提取图像的亮度信息,而且进行图像增强的同时能有效的抑制图像的噪声,且对于不同的雾天天气具有良好的自适应性。3、在研究现有压缩图像增强算法的基础上,将Retinex理论引入到JPEG2000压缩框架中,达到增强JPEG2000压缩图像的目的,具体做法分为两个步骤：Retinex增强和改进亮度量化表。首先以Retinex理论为基础,将小波变换的低频系数看作入射光分量,高频系数看作反射光分量,通过修改低频系数来调整场景光照,修改高频系数来提高图像的对比度；然后通过判断各子块的活动性,自适应来修改亮度量化表,达到可以保留更多的细节,抑制块状效应的目的。更多还原

【Abstract】 Image enhancement as an important pre-processing means has gained much attention by many people because of its widely used. Image enhancement is purposefully emphasized the gloable or local image pixel value features to improve the visual effect or to meet the specific analysis. Retinex theory and related algorithm is new idea of image enhancement.Retinex method is theoretically based on the color constancy of human visual system. The basic strategy is simulating the process of human vision to enhance image contrast and resume true color of object.This paper proposed a new Retinex algorithm and JPEG2000 compression framework by analysising the traditional Retinex algorithm and wavelet multi-resolution features. Secondly a new method to enhance JPEG2000 compressed image is proposed, in which Retinex theory is introduced into JPEG2000 compressed frame.The main content of this thesis is shown as follow:1.Based on the view of selecting the path of pixels comparison, Common Computational Strategies of Retinex algorithm has been analysised, including Retinex algorithm based on random paths, the McCann’s Retinex algorithm based on iterative calculation, the Retinex algorithm based on center/surround, Then this paper pointed out characteristics and limitations of Retinex algorithm, and verified in experiments.2.Because of the problem that the traditional enhancement algorithm can not suppress the noise effectively when enhanced the image. This paper proposes a novel algorithm of estimating the image illumination and noise in the wavelet domain based on Retinex theory. By using the feature of the wavelet multi-resolution, algorithm suppress the volatility of the low-frequency coefficient and use the model of multi-subband independent coefficient to estimate the high-frequency coefficient, at the same time using soft-hard estimated method to suppress the image noise. The experimental results show that the algorithm can extract the image illuminate vector effectively, but also can suppress the image noise and adapt in different foggy weather.3.In the base of recently research for image enhancement algorithm in the compressed domain, we introduce the Retinex theory into the JPEG2000 code process to achievie the goal for enhancement the JPEG2000 compression image. The procedure may divide into two steps. First, using the Retinex algorithm to carry on the enhancement; Secondly,improving the Luminance quantification table.It take the Retinex theory as the foundation,regarding the DC coefficient of DWT coefficient as the incident light component and the AC coefficient of DWT coefficient as the reflected light component.the goal for revises the DC coefficient is to adjust the scene illumination condition and for the AC coefficient is to enhance the image contrast gradient,then judging the activity of the great block to adjust the luminance quantification table adaptively to achieve the goal of preserving enhancement detail and suppressing the massive effect.更多还原