计算机工程

• •    

结合修剪均值与高斯加权中值的图像去噪

  

  • 发布日期:2020-12-25

Image denoising algorithm based on trimmed mean and Gaussian weighted median

  • Published:2020-12-25

摘要: 为克服现有的脉冲噪声去除算法的不足,进一步提升去噪的鲁棒性和有效性,提出 了一种结合修剪均值与高斯加权中值的图像去噪方法。先根据脉冲噪声与信号像素在分布和 灰度上的特征,以局部统计方式进行噪声检测;将取最小灰度或最大灰度且与邻域像素相关 性较小的像素识别为噪声。然后,对平滑区域和细节区域中的噪声,分别用自适应的修剪均 值和自适应的高斯加权中值进行去噪处理,以高斯加权算子反映邻域像素之间的影响和相关 性。实验结果表明,所提出的算法在视觉效果、PSNR 和 SSIM 以及计算速度上均优于现有 算法,在彻底去除噪声的同时,较好地保持图像的边缘和细节结构。

Abstract: In order to overcome the inefficiency of existing filters in impulse noise removal, and achieve higher robustness and better effectiveness than the existing filters, an image denoising scheme based on trimmed mean and Gaussian weighted median is proposed. By exploiting the features of impulse noise and noise free pixels in distribution and intensity, the noisy pixels are first discriminated from noise free ones in a local statistical way, the pixels with minimum or maximum intensity and with less correlation with neighboring pixels, are taken as noisy pixels. And then, the noises in smooth region and detail region, are processed respectively by adaptive trimmed mean filter and adaptive Gaussian weighted median filter, the impact and correlation of neighboring pixels are presented by Gaussian weighted operator. Experimental results show that in terms of visual quality, PSNR and SSIM, and computational efficiency, the proposed method is superior to existing filters, preserves better the edges and detail structures of image while removes the noise thoroughly.