采用双树复小波和混合概率模型的光学相干层析图像去噪

摘要

To remove speckle noise in optical coherence tomography (OCT) images, the ProbShrink algorithm based on the dual-tree complex wavelet transform and a mixed probability model is employed. After studying the signal and noise distribution in OCT images, a mixed probability model in microscopic-level is introduced. Logarithm of the OCT image is first decomposed using dual-tree complex wavelet transform. The coefficients consistent with edges obey the generalized Gaussian distribution, while others obey the Gaussian distribution. An improved ProbShrink algorithm is used to shrink the wavelet coefficients. Experiments show that this method can significantly improve signal-to-noise ratio while hold edge preservation index relatively steady. The performance is better than that of traditional wavelet based OCT image denoising methods.%为了去除光学相干层析图像中的斑点噪声,提出了基于双树复小波变换的混合概率模型ProbShrink算法.针对原始光学相干层析图像中信号和噪声的分布特点,在微观层面引入了混合概率模型:将OCT图像取对数后进行双树复小波变换,对于层状边缘中与边缘点“方向一致”的小波系数,采用广义高斯模型描述；对于其他小波系数,则采用高斯模型进行描述.而后采用改进的ProbShrink算法进行去噪.实验结果表明,该算法在大幅提升信噪比的情况下保持边缘锐度的相对稳定,优于传统的基于小波变换的去噪方法.