双树复数小波变换的视网膜图像半盲解卷积复原

Retinal image semi-blind deconvolution restoration based on dual tree complex transform

针对导致自适应光学视网膜图像降质退化的原因，提出了一种结合双树复数小波变换（DT—CWT）和图像半盲解卷积复原算法的方法。首先，对经过自适应光学实时校正技术得到的视网膜图像进行DT—CWT分解，得到低频和高频部分对应的图像。将自适应光学成像系统中残余像差重建的光学传递函数作为图像复原模型的初始估计点扩散函数（PSF），并对低频部分图像进行条件约束的迭代半盲解卷积复原；对高频部分的图像进行去噪处理。最后，将处理后的高频和低频部分图像进行双树复数小波逆变换，获得复原图像。实验和结果表明：由该方法处理的视网膜细胞图像质量得到明显提高，图像客观质量评价参数相对于原始图像提高了5倍多；在视网膜细胞的空间频率范围内（70～90（°）-1），复原图像功率谱平均值提高了5倍左右，有助于对视网膜细胞的高分辨率观察。

A combination of dual tree complex transform and the image processing algorithm of semi blind deconvolution was proposed to eliminate the factors which make the image worse. Firstly, the retinal image obtained from the adaptive optics system was decomposed with dual tree complex transform into two parts. The low frequency image was processed by the algo- rithm of semi-blind deconvolution with some constraints. And the optical transfer function was used as initial parameter estimate which was constructed with the residual aberration of image system. The high frequency image was processed by denoising. Fi nally, the goal image was obtained by combining the processed images. The experimental results show that the retinal image qual ity is improved with this method, the image objective quality evaluation parameters are increased more than 5 times compared with the original image, and the average power spectrum is improved about 6 times in the spatial frequency range（70~90（°） -1） of retinal cells, which show that the method can contribute to satisfying the requirement of observation of human retinal image.