光学 精密工程, 2020, 28 (1): 244, 网络出版: 2020-03-25
基于Anscombe变换的X射线图像序列盲源分离降噪
X-ray image denoising using blind source separation in anscombe domain
X射线图像 泊松噪声 Anscombe变换 非线性主分量分析 盲源分离 X-ray image poisson noise anscombe transform nonlinear principal component analysis blind source separation
摘要
为降低泊松噪声对X射线图像质量的影响, 本研究提出一种采用非线性主分量分析(NLPCA)对X射线图像序列进行盲源分离的降噪方法。该降噪方法首先采样一序列X射线图像, 并通过Anscombe变换将图像中泊松噪声转化为高斯噪声; 然后将每张含噪声图像视为噪声分量和信号分量的组合, 进而采用NLPCA将信号分量和噪声分量分离达到降噪目的; 最后通过Anscombe逆变换获取最终降噪图像。研究结果表明: 当序列中含噪声图像张数从2增加到50时, 提出的降噪方法可以将Shepp-Logan头模型含噪声图像的PSNR值由28.289 4 dB提高到37.267 8 dB、SSIM值由0.700 7提高到0.963 8。相比较常用的降噪算法, 提出的降噪方法在有效消除X射线图像中泊松噪声的同时, 使图像中细节轮廓保留更完整。
Abstract
To remove the Poisson noise from the X-ray images, in this paper, it was proposed that noise was reduced by using Nonlinear Principal Component Analysis (NLPCA) from the X-ray image sequence. At first, an X-ray image sequence was sampled and the Poisson noise in images was converted into Gaussian noise through Anscombe transform; every noisy image was regarded as a combination of the noise components and the signal component, and then NLPCA was used to separate the signal component from the noise components to reduce noise; the final denoised image was obtained by using Anscombe inverse transform. The results show that, when the number of noisy images in the sequence increases from 2 to 50, the proposed denoising method increases the noisy Shepp-Logan image′s PSNR value from 28.289 4 dB to 37.267 8 dB and increases the SSIM value from 0.700 7 to 0.963 8. Compared with other denoising methods, the proposed denoising method can preserve more image details while reducing the Poisson noise.
沈帆, 李翰林, 孙斌, 喻春雨. 基于Anscombe变换的X射线图像序列盲源分离降噪[J]. 光学 精密工程, 2020, 28(1): 244. SHEN Fan, LI Han-lin, SUN Bin, YU Chun-yu. X-ray image denoising using blind source separation in anscombe domain[J]. Optics and Precision Engineering, 2020, 28(1): 244.