基于压缩感知的高密度分子定位算法比较

Comparison of Algorithms of High-Density Molecule Localization Based on Compressed Sensing

为了提高荧光超分辨显微技术的时间分辨率,提出了各种高密度荧光分子定位算法。对基于压缩感知模型的凸优化(CVX)工具箱的内点算法、同伦算法以及正交匹配追踪(OMP)算法的重构密度、定位精度、定位时间进行比较。模拟结果表明,CVX方法和同伦算法能够在高密度情况下实现精确定位,OMP算法与同伦算法运行时间比CVX算法短,但OMP算法在高密度的情况下定位精度相对其他2种算法明显下降。实验结果表明,3种算法都能实现高密度的荧光分子定位,CVX方法和同伦算法具有较好的重构效果;在500幅图像重构中,同伦算法与OMP算法的速度相比于CVX算法分别提高了14.9倍和21.2倍,大幅度缩短了重构时间。

In order to improve the time resolution of super-resolution fluorescent microscopy, the methods of high- density molecule localization have been proposed. Three algorithms based on compressed sensing models, including the interior-point method in the CVX toolbox, the homotopy method, and the orthogonal matching pursuit （OMP） algorithm, are investigated. We compare the identified density, localization precision, and execution time by using these algorithms in the simulations and experiments. Simulation results show that the CVX and homotopy methods can accurately locate in the high molecule density, but the CVX method has the longest running time among these methods. The OMP method has low localization precision in the high density. The experimental results show that these algorithms can realize the localization of high molecule density. The CVX and homotopy methods get better results than OMP method in the localization precision. For the localization of 500 images, the homotopy and OMP methods are 14.9-fold and 21.2-fold faster than CVX method, which can greatly shorten the reconstruction time.