Spectral calculation model for machine vision image enhancement

Abstract

With the development of artificial intelligence technology, the demand for image quality in machine vision systems is increasing. However, current research mainly focuses on chromaticity and light environment indicators, and cannot fundamentally solve the problem. To solve this problem, we established a machine vision optimal spectral calculation model from the perspective of physical energy, designed a narrowband spectral experiment, and analyzed it using JS divergence. The results showed that the calculated optimal spectrum significantly improved the image brightness and JS divergence compared to Standard White, with a maximum increase of 135.66% in image brightness and 82% in JS divergence. Research has found a significant linear correlation between the brightness value of machine vision images and the irradiance with a coefficient of 1, but not with the illumination. It was also found that the divergence of JS is not related to the irradiance, but has a significant linear correlation with the difference in spectral distribution with a coefficient of 1. These findings will provide a new basis and ideas for the light environment design of machine vision systems, provide new methods for improving system image quality, and have a significant positive impact on deep learning of the machine vision system.