光学 精密工程, 2020, 28 (1): 189, 网络出版: 2020-03-25
基于压缩感知的频域OCT图像稀疏重构
Sparse reconstruction of frequency domain OCT image based on compressed sensing
图像重构 光学相干层析扫描 压缩感知 采样率 image reconstruction optical coherence tomography compressed sensing sampling rate
摘要
为了减轻频域光学相干断层扫描成像(Frequency Domain Optical Coherence Tomography, FD-OCT)中高数据量导致的后续数据采集与处理系统的压力, 同时解决成像时间和成像质量之间的矛盾, 引入了压缩感知技术, 并对该技术中的重构算法进行了重点研究。首先, 通过分析压缩感知技术的框架, 利用离散余弦变换(Discrete Cosine Transform, DCT)获得频域OCT图像的稀疏表示。接着, 利用高斯随机矩阵对OCT图像进行线性观测。然后, 研究了FOCUSS(Focal Underdetermined System Solver)重构算法的原理, 并在算法中结合分块思想、引入正则项lp范数以及嵌入各向异性平滑算子。最后, 组合所有小图像块, 得到整幅频域OCT图像的压缩感知重构结果。实验结果表明: 改进重构算法的运行时间由78.65 s缩短为1.89 s, 并且显著改善了图像块效应, 将重构图像的PSNR值提高了1.6~2.7 dB, SSIM值可达到0.938 3。压缩感知技术可以用较小的采样数据量精确重构出原始频域OCT图像, 改进FOCUSS重构算法可以在一定程度上实现频域OCT图像重构效率和重构质量的平衡。
Abstract
In order to alleviate the pressure of subsequent data acquisition and processing systems caused by high data volume in Frequency Domain Optical Coherence Tomography (FD-OCT), and to address the contradiction between imaging time and imaging quality, we introduced compressed sensing technology and focus on the reconstruction algorithm in this technology. First, we analyzed the framework of the compressed sensing technology, the frequency domain OCT image was sparsely represented by Discrete Cosine Transform. Next, we used Gaussian random matrices to perform linear observations on OCT images. Then, we studied the principle of FOCUSS (Focal Underdetermined System Solver) reconstruction algorithm, and combined the block idea, introduced the regular term lp norm and embed anisotropic smoothing operator in the algorithm. Finally, we combined all the small image blocks to obtain the compressed sensing reconstruction result of the whole frequency domain OCT image. Experimental results indicate that the running time of the improved reconstruction algorithm is shortened from 78.65 s to 1.89 s, and the image block effect is significantly improved, the PSNR value of the reconstructed image is improved by 1.6-2.7 dB, and the SSIM value can reach 0.938 3. Compressed sensing technology can accurately reconstruct the original frequency domain OCT image with a small amount of sampled data. The improved FOCUSS reconstruction algorithm can achieve the balance of frequency domain OCT image reconstruction efficiency and reconstruction quality to some extent.
陈明惠, 王帆, 张晨曦, 李福刚, 郑刚. 基于压缩感知的频域OCT图像稀疏重构[J]. 光学 精密工程, 2020, 28(1): 189. CHEN Ming-hui, WANG Fan, ZHANG Chen-xi, LI Fu-gang, ZHENG Gang. Sparse reconstruction of frequency domain OCT image based on compressed sensing[J]. Optics and Precision Engineering, 2020, 28(1): 189.