Optical quantitative phase imaging (QPI) is a frequently used technique to recover biological cells with high contrast in biology and life science for cell detection and analysis. However, the quantitative phase information is difficult to directly obtain with traditional optical microscopy. In addition, there are trade-offs between the parameters of traditional optical microscopes. Generally, a higher resolution results in a smaller field of view (FOV) and narrower depth of field (DOF). To overcome these drawbacks, we report a novel semi-supervised deep learning-based hybrid network framework, termed ContransGAN, which can be used in traditional optical microscopes with different magnifications to obtain high-quality quantitative phase images. This network framework uses a combination of convolutional operation and multiheaded self-attention mechanism to improve feature extraction, and only needs a few unpaired microscopic images to train. The ContransGAN retains the ability of the convolutional neural network (CNN) to extract local features and borrows the ability of the Swin-Transformer network to extract global features. The trained network can output the quantitative phase images, which are similar to those restored by the transport of intensity equation (TIE) under high-power microscopes, according to the amplitude images obtained by low-power microscopes. Biological and abiotic specimens were tested. The experiments show that the proposed deep learning algorithm is suitable for microscopic images with different resolutions and FOVs. Accurate and quick reconstruction of the corresponding high-resolution (HR) phase images from low-resolution (LR) bright-field microscopic intensity images was realized, which were obtained under traditional optical microscopes with different magnifications.
Perfect optical vortices (POVs), consists of a single bright ring structure, has been widely studied owing to its radius independent of orbital angular momentum (OAM). However, most of the existing works about POVs are limited to single ring structure. Flexible shaping of intensity distribution of POVs is vital for multiple applications. In this paper, we propose a method generate phase tunable multi-ring perfect optical vortices (MR-POVs) where each ring size is independent of its OAM. The scheme is based on the radical discontinuous spiral phase plate (RD-SPP) which introduces controllable phase jumps along radial direction. It is experimentally demonstrated that the vortex nature of the MR-POVs through an interferometric method, showing that each ring of MR-POVs possesses same topological charge value (magnitude and sign), and the intensity ratio between each ring can be freely regulated by adjusting phase distribution, which could offer more flexible optical gradient force for guiding and transporting particles. In addition, simulation and experimental results show that the integer and fractional MR-POV can generated by the independent regulation of angular and radial factors. This work expands our understanding of multi-ring POV and may provide a new idea for optical tweezers and OAM communications.
Gaze tracking is an assistant system of human-computer interaction. Aiming at the problem of high misjudgment rate and long time-consuming of traditional iris location methods, this paper proposes a gaze tracking method based on human eye geometric characteristics to improve the tracking accuracy in 2D environment. Firstly, the human face is located by face location algorithm and the position of human eye is estimated roughly. Then the iris template is built by iris image, and the iris center location algorithm is used to locate the iris center position. Finally, the eyes corners and iris center points are extracted to locate the eye area accurately and obtain the binocular image. The binocular images are input into the feature extraction network as multi-modal information in parallel, and the convoluted feature channels are reconstructed using the weight redistribution module in the network. Then the reconstructed features are fused in the full connection layer. Finally, the output layer is used to classify the reconstructed features. Experiments were carried out on a self-built screen block data set. For 12 classified data, the lowest recognition error rate is 5.34%.