自适应非整数步长的分数阶微分掩模的图像纹理增强算法  

An Algorithm for Texture Enhancement Based on Fractional Differential Mask Using Adaptive Non-integer Step

作　　者：姒绍辉[1] 胡伏原[1,2] 付保川[1] 李金祥 

机构地区：[1]苏州科技学院电子学院,苏州215011 [2]江苏省现代化企业信息化应用支撑软件工程研发中心,苏州215011

出　　处：《计算机辅助设计与图形学学报》2014年第9期1438-1449,共12页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金重点项目(61231016);江苏省自然基金(BK2012166);江苏省建设厅科技项目(JH21);江苏省高校自然科学基金(12KJB510031);江苏省研究生培养创新工程基金(CXZZ13_0854)

摘　　要：图像纹理增强是计算机图形学、计算机视觉和模式识别等领域里的一个重要问题.通过分析分数阶微分原理和纹理图像的特性,提出一种自适应非整数步长的分数阶微分掩模算法,并将其应用于纹理图像增强中.利用图像纹理间的高度自相关性自适应地构建局部不规则的自相关掩模区域,剔除相关性较低的像素并降低噪声干扰;同时,突破传统分数阶微分数值计算采用单位步长的思想,分析不规则掩模区域的臂长特征,自适应地估计非整数步长;最后建立局部线性模型实现对非整数步长处的像素灰度值的准确估计,提高分数阶微分数值解的逼近程度.实验结果表明,该算法能够提高分数阶微分解析值的精确度,有效地增强了图像平滑区域中的复杂纹理细节.Image texture enhancement is an important issue in computer graphics, computer vision and pattern recognition. By analyzing fractional differential principle and characteristics of texture images, a fractional differential mask with self-adaptive non-integer step is proposed to enhance texture images. Firstly, a non-regular autocorrelation mask region is adaptively constructed by taking advantage of the high correlation among textures in images to remove lowly correlated pixels and reduce the interference from noises. Afterwards, breaking through the traditional method in which the step in the numerical calculation has to be taken in unit size, pixels spanning on the arms of autocorrelation region are divided adaptively based on arms＇ length to further improve the accuracy of fractional differential computation. Finally, local linear modeling is adopted to estimate gray scale value for non-integer pixels. Experimental results show that the proposed algorithm can significantly improve the precision of numerical analysis of fractional differential computation, and thus can further enhance the complex texture details in smooth areas of images.

关 键 词：纹理增强 分数阶微分 非整数步长 线性估计 

分 类 号：TN911.73[电子电信—通信与信息系统]