A Rapid Training Method for Color Correction Models in Machine Vision Imaging

When using machine vision systems to replace human eyes for color difference detection of products, machine vision systems are greatly affected by hardware and environmental conditions. When one or more factors change, it often reduces the accuracy of machine vision systems for color difference detection. The traditional sample collection method has a complex process and takes a long time. If samples are customized based on the hardware configuration and environmental conditions of the system, the cost is high and the generalization is poor. To address these issues, this article proposes a rapid training scheme. This scheme adds a compensation model to the basic model of color correction in the machine vision system, uses polynomial regression to correct the collected colors, and tests the results. The experimental results show that after changes in device and environmental conditions, the accuracy and performance of the rapid training color correction model are higher than those of the Initial color correction model.

I. Introduction

After the 21st century, the competition in the domestic printing industry has become more intense. With the continuous improvement of printing processes and product design, and the wider use of raw materials, a large number of exquisite printing products have emerged in the market. However, products with color differences are difficult to win favor due to their lack of precision. Therefore, color difference detection has become an essential part of industrial production [1]. Traditional manual detection methods have low accuracy, slow speed, and are more susceptible to subjective influence, gradually being replaced by intelligent color difference detection methods based on machine vision [2].