改进型多项式匹配图像去噪算法的研究与应用  

Application and research of improving polynomial fitting algorithms for images denoising

作　　者：陈波 鄢爱兰[2] Chen Bo;Yan Ailan(The Second Hospital,University of South China,Hengyang 421001,China;Academic Affairs Department,University of South China,Hengyang 421001,China)

机构地区：[1]南华大学附属第二医院计算机中心,衡阳421001 [2]南华大学教务部,衡阳421001

出　　处：《电子测量与仪器学报》2021年第4期177-186,共10页Journal of Electronic Measurement and Instrumentation

基　　金：湖南省教育厅优秀青年项目(15B207)资助。

摘　　要：为进一步改善常规多项式匹配算法的去噪能力,针对常规多项式匹配算法容易模糊图像边缘/纹理等细节问题,提出一种基于边缘保护的改进多项式匹配滤波的算法。该方法在常规多项式匹配算法基础上,改进滤波窗口的选择方式,沿着图像纹理走向方向提取自适应滑动滤波窗口,选择具有最小匹配误差的窗口进行匹配滤波并作为最终输出结果。然后分别在灰度图像和实际CT图像进行测试,经数据验证表明,该方法在有效压制噪声的前提下能够保持边缘/纹理信息,峰值信噪比平均提升80%以上,均方根误差减少80%以上。认为和常规多项式滤波方法、中值滤波方法、双边滤波方法、边缘保持滤波方法、三维块匹配去噪算法和去噪卷积神经网络方法相比,改进方法能够有效提升图像视觉效果,满足图像应用要求,具有良好的应用前景。In order to further improve the de-noising ability of conventional polynomial matching algorithms,an improved polynomial matching filtering algorithm based on edge protection is proposed,aiming at the details such as image edge/texture easily blurred by conventional polynomial matching algorithms.Based on the conventional polynomial matching algorithm,it improves the selection method of filtering window,extracts adaptive sliding filtering window along the direction of image texture direction,selects the window with minimum matching error for matching filtering and takes it as the final output result.Then the grayscale image and the actual CT image were tested respectively.The data verification shows that this method can maintain edge/texture information while effectively suppressing the quantization of noise,and the peak signal-to-noise ratio is increased more than 80%,and the root mean square error is reduced more than 80%.Compared with the conventional polynomial filtering method,median filtering method,bilateral filtering method,edge preserving filtering method,BM3 D and DnCNN method,the improved method can effectively improve the image visual effect,meet the requirements of image application,and has a good application prospect.

关 键 词：边缘保护 纹理方向 多项式匹配 自适应滑动滤波窗口 压制噪声 

分 类 号：TN911.73[电子电信—通信与信息系统]