Fig. 5. Bio-inspired gate-tunable scotopic-photopic adaptation and in-sensor image enhancement.
a Schematic of the human eye adjusting to bright (cone-dominated) and weak (rod-dominated) light conditions. b Schematic illustration of the response intensity of cone and rod cells as a function of light intensity. c Normalized photocurrent (Iph) as a function of incident light power density (Pin) at Vgs = -4 V and Vgs = 0 V, where the data points represent the experimental measurements, and the dashed lines represent the corresponding fitting curves. d Image enhancement demonstration. Synthetic input images are algorithmically processed to show linear imaging under bright light and nonlinear enhancement under weak light, based on the experimentally measured characteristics of the OSDs. e Convolutional neural network (CNN)-based recognition workflow using weak-light and enhanced datasets. f Recognition accuracy of weak-light and enhanced CIFAR-10 datasets as a function of training epochs.