一种2D-VMD与NLM结合的芯片图像去噪算法研究

Chip image denoising algorithm based bi-dimensional variational mode and non-local mean filtering

摘要:

为了提高智能电表芯片图像的字符识别精度，需要消除芯片图像中的噪声，以减小干扰。文章提出了一种基于二维变分模态分解算法(two dimensional variational mode decomposition，2D-VMD)与非局部均值(non-local mean, NLM)滤波的芯片图像去噪算法。首先利用2D-VMD将含有噪声信号的芯片图像分解为K个模态分量；然后根据提出的结构相似(SSIM)阈值设置方法确定噪声分量并将其去除，使用剩余的有效分量重构图像；最后通过非局部均值滤波算法对重构后的图像进行处理，进一步滤除残余噪声，达到二次去噪的效果。实验结果表明，相比传统的图像去噪算法，提出的算法能在较好保留原始芯片图像的字符信息的基础上，去除不相关的噪声干扰，使去噪后的芯片图像的均方误差值变小，峰值信噪比增大，提高芯片图像质量。

Abstract:

To retain the character information of the smart meter chip image and remove noise interference, a chip image de-noising algorithm combining two-dimensional variational modal decomposition algorithm and non-local mean fil-tering was proposed. On the basis of the center frequency observation method, 2D-VMD was used to decompose the noisy chip image into K modal components, and the noise modal components were removed according to the structural similarity measure (SSIM) threshold method, the NLM algorithm used to the effective component in the 2D-VMD algorithm and through the way of the secondary image denoising to achieve the effect of precise denoising. Experimental results show that the mean square error value and the peak signal-to-noise ratio of proposed algorithm are superior to the traditional filtering algorithms.