A Bio-Optical Sensor Based on Bacteriorhodopsin for Self-Adaptive Image Denoising

Image denoising plays a crucial role in enhancing image recognition accuracy. Image denoising is mainly solved by software level in traditional methods, which often face limitations such as loss of image details and applicable to specific data sets, especially under high-noise conditions. To address these limitations, hardware-based denoising methods have emerged, primarily enhancing target regions based on device-specific photosensitive characteristics. In this work, a bio-optical sensor based on bacteriorhodopsin for self-adaptive image denoising is proposed. By using the unique photoelectric characteristics of bacteriorhodopsin, the photocurrent of the bio-optical sensor gradually decreases over time. More surprisingly, the photocurrent duration time is affected by the intensity of the light. The greater the light intensity, the longer the photocurrent duration time. By using this photoelectric property, the target regions and the noise regions in the image captured by the bio-optical sensor can achieve enhanced contrast, thereby achieving self-adaptive image denoising. A simulation was performed to evaluate the image denoising effect of the bio-optical sensor when processing different data sets (the MNIST and fashion-MNIST). The results show that, after the image denoising processing of the simulated sensor array, the recognition rate in the neural network has improved for the two different data sets (Fashion-MNIST: 3.23%, MNIST: 2.8%), which indicates that the proposed hardware-based denoising method is effective. This paper offers a novel bio-optical sensor with hardware-level self-adaptive image denoising capability, which demonstrates strong potential for applications in future intelligent machine vision.