视网膜图像的解卷积方法研究

Deconvolution algorithm of retinal images

基于自适应光学的激光共焦扫描检眼镜(AOSLO)能够利用自适应光学校正眼底像差,从而对活体视网膜实现高分辨率成像。但是由于波前探测误差、变形境自由度和行程等硬件上的限制,眼底像差不可能完全被校正,并且信号探测器光电倍增管(PMT)存在较大的噪声,因此可以用解卷积的方法进一步提高图像质量。本文改进了增量维纳滤波算法,将迭代步长减为原来的1/5,使迭代不会偏离初始值太远,而增量维纳滤波迭代初始值是由系统中的波前传感器得到相对准确的值。采用上述方法对AOSLO采集到的视网膜细胞层和血管层图像做解卷积,并复原了整个系统的点扩散函数(PSF)。由图像功率谱表明,增量维纳滤波算法能有效去除图像噪声,增加图像对比度。

Adaptive optics can be used to compensate for the wave aberration of human eyes,and high resolution imaging of living human retina is achieved by adaptive optics confocal scanning laser ophthalmoscope（AOSLO）.Because the correction is limited by the hardware and a lot of noises generated by photomultiplier tube（PMT）,a deconvolution technique can be used to improve the image quality.The incremental Wiener filter is improved by decreasing the iterative steps to 1/5 of the original number.And it can be convergent near the initial value which is calculated from the wavefront measurement.The method above is used to deconvolute retinal blood vessel as well as cell layer images,and revise point spread function（PSF）.The power spectra indicate that the incremental Wiener filter can remove the noise and enhance the image contrast.