基于自适应分数阶微分的引导滤波及其应用  

Guided image filter and application based on adaptive fractional-order differential

作　　者：谢伟[1] 游敏[1] 周玉钦[1] 

机构地区：[1]华中师范大学计算机学院,武汉430079

出　　处：《计算机应用研究》2017年第1期283-286,301,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61501198;41101425;41201413);湖北省自然科学基金面上资助项目(2014CFB461);武汉市青年科技晨光计划资助项目(2014072704011248);华中师范大学中央高校基本科研业务费资助项目(CCNU14A05017)

摘　　要：引导滤波算法具有保边平滑的功能,但传统引导滤波方法容易导致图像平滑区域过度模糊、细节丢失的问题。为了使引导滤波在保持高频信息的同时结构化输出低频灰度,提出了一种基于自适应分数阶微分的引导滤波算法。以分数阶微分理论为基础定义了分数阶微分掩模,并结合图像梯度、二维信息熵和局部方差权值构造了自适应分数阶微分阶数函数来有效检测图像纹理和梯度变化,从而将图像局部特性转移到引导图像中,确保在平滑去噪的同时保持图像纹理细节。实验结果表明,算法具有良好的边缘和纹理保持特性,将算法运用到基于PCA和SVM的人脸识别图像预处理中,能一定程度地提升人脸识别率。Guided image filter has the function of edge preserving and smoothing, but the traditional guided image filter can lead to details losing and blurring. In order to keep high-frequency information and compute low-frequency structurally, this paper proposed guided image filter based on adaptive fractional-order differentiation. Firstly, it defined the fractional-order differential mask based on the fractional-order differential theory. Secondly, it designed fractional-order differential order function according to the image gradient, local information entropy and local variance, which could ＇detecte the changes of texture and gradient. Then,it transferred local conditions to the guided image that ensured not only preserving egdes and denosing, but also keeping texture. The experiment results show, the proposed algorithm is skilled at preserving egdes and texture, and it can improve the accuracy of face recognition based on PCA and SVM in image preprocessing.

关 键 词：保边平滑 分数阶微分 掩模 局部方差 人脸识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]