信噪局部方差自适应的小波阈值滤波

Wavelet threshold filtering based on adaptive local variance of signal and noise

为了在去除高斯噪声的同时更有效地保持图像的边缘和细节,提出了信噪局部方差自适应的小波滤波方法.根据图像与高斯噪声的小波系数的分布特征,提出了一种信噪局部方差自适应的阈值.同时,鉴于无噪图像的小波系数具有平滑连贯性,提出一种连续的、可微的且无限逼近原小波系数的阈值函数.阈值依据信噪强度对信号系数与噪声系数进行区分,阈值函数依据阈值对小波系数进行量化处理,以去除噪声.实验结果表明,所提出的方法对图像去噪所得的PSNR(peak signal to noise ratio)和SSIM(structural similarity index)值以及图像的视觉效果,相对于现有的小波去噪方法有较大的提升,在彻底去除高斯噪声同时,更有效地保持图像的边缘和细节.

Aiming at removing Gaussian noise and maintaining the edge and detail of the image more effectively,a wavelet filtering method based on adaptive local variance of signal and noise was proposed.In view of the distribution characteristics of wavelet coefficients of the image and Gaussian noise,this method proposed an adaptive threshold based on local variance of signal and noise.And,in view of the smooth coherence of wavelet coefficients of noiseless images,a continuous and differentiable threshold function which infinitely approximates the original wavelet coefficients was introduced.The threshold distinguished the coefficients of noise from that of image with the intensity of signal and noise,and the threshold function processed the wavelet coefficients quantitatively,with the aim to remove noise.Experimental results showed that compared with the existing wavelet denoising methods,the PSNR(peak signal to noise ratio) and SSIM(structural similarity index) values and the visual quality of denoised image derived by the proposed method increase greatly,the proposed method removes the Gaussian noise completely,and maintains the edge and detail of the image more effectively.