An approach for recovering the phase information from the detected intensity was proposed in this work. Unlike the conventional approach based on the Gerchberg-Saxton algorithm, the proposed approach recovered the phase information via an alternative technique in the realm of deep learning, the residual neural network. The database we utilized to train the network was collected by a Michelson-based interferometer, where a spatial light modulator was implemented to provide the phase modulation as the phase object. As the result, the mean absolute error of each pixel was 0.0614π.
Organic light emitting diode generally has serious non-uniformity phenomena due to the instability of organic processing, called Mura. In this paper, we propose an automatic Mura detection model to mimic the human perception and detect Mura pixel-wisely. First, we extract regions of interest from the original image with different sizes of windows, and then we verify these regions by SEMU criterion. Consequently, we implement human visual properties based on the contrast sensitivity function filtering and ModelFest matching to segment Mura regions. As the result, our approach can successfully detect Mura with various sizes and shapes, which could have a great impact on the display industry.
Eccentric infrared photorefraction is an attractive vision screening method which is widely used for uncooperative subjects, such as infants and toddlers. Unlike conventional slope-based photorefraction, a deep neural network is used to predict refractive error in this study. Total 1216 ocular image were collected by a homemade photorefraction device, whose corresponding refractive error was measured by a commercial autorefractor device, to create a series of dataset for our deep neural network. The mean squared error of the preliminary result is ±0.9 diopter, which indicates its feasibility and can be improved with bigger database.