A Method of Biomimetic Visual Perception and Image Reconstruction Based on Pulse Sequence of Events

Abstract

A method of biomimetic visual perception and image reconstruction based on pulse sequence of events is proposed, and the proposed method facilitates a behavioral model of vision image sensing. Based on the model, the relationship between light condition and quality of reconstructed videos is analyzed. In this sensing method, pulse sequence is defined as synchronous pulse sequence. The sensing method adopts a synchronous pulse sequence readout and asynchronous pixel level sensing mechanism. Light intensity can be calculated by pulse sequences and image or video can be further reconstructed. As the photocurrent increases, the real-time performance of reconstructed image increases, but the depth of the reconstructed image decreases. Experiments are performed on sensing models with two types of pulse sequence generating methods. Method 1 adopts 1-bit data for each pixel, and Method 2 adopts 1-bit data accompanied by 3-bit time label which enables a lower frame rate. In the experiment with static images, the reconstructed images show better image depth under lower photocurrent. The MSE (mean square error) of the reconstructed image with 450fA maximum photocurrent on Method 1 is 0.0737, whereas the MSE in the same condition on Method 2 is 0.0083. In the experiment with high-speed videos of 20000fps, the reconstructed videos with various photocurrents are evaluated by MSE analysis. When maximum photocurrent is between 6750fA and 8250fA as well as between 11250fA and 16500fA, the reconstructed high-speed videos have good MSE results, where video real-time and image depth are balanced.