【论文阅读】3D Deconvolution with Deep Learning

3D Deconvolution with Deep Learning

主要

感觉像毕业论文一样的
比较了四种解卷积方法，两种是经典方法，两种是深度学习方法（SRCNN和Isonet）。
比较他们的效果。
SRCNN是超分辨网络。。。

没有提出新的方法。知识结果比较。

他们的数据集貌似挺合适以后项目的。
The dataset used for all experiments is taken from Cell Imaging Library which is an open-source repository for images of biological specimen. The dataset [2] shows microtubules in a Drosophila S2 cell. It contains a stack of 44 wide-field (WF) images and 44 structured-illumination (SIM) images of size 1904x1900 which makes the volume size to be 1904x1900x44. Due to limitations of computing power and ease of iterating, the images were resized in the lateral directions. Some basic image thresholding and corrections were made with exisiting pipelines in Fiji (ImageJ) [5]. It is a reasonable assumption to make that the SIM images are the ground truth and the the WF images are blurred observed data.