Abstract:

Objective In the textile industry, fabric color is a critical factor affecting the final product's appearance and quality. With the dynamic shifts in fashion trends and increasing diversity in consumer demands, the production pace in the textile sector has accelerated, necessitating more diverse and rapidly updated color options. Consequently, developing a convenient and efficient textile color measurement system is vital to meet the industry's evolving demands. This system is designed for scenarios such as offline trading where traditional measurement system may not be feasible, thereby facilitating accurate color assessment and transactional decisions in the textile domain.

Method In this research, a portable fabric image color measurement system compatible with multiple devices was developed, which comprises a foldable, enclosed image acquisition device and an algorithm for color measurement. The design of image acquisition device incorporated a detachable and collapsible structure, enhancing the system's portability. To ensure accurate color measurement, the study integrated polynomial regression and K-Means clustering algorithms to devise a categorical approach to color measurement. This method involved applying different polynomial regression models based on the specific color categories of the fabrics, thereby facilitating precise color measurement.

Results The process began with capturing images of Datacolor SpyderCheckr24 color swatches using a smartphone. Subsequently, these images underwent processing through a classification correction algorithm developed in this study. The analysis of color differences after correction indicated that the performance of this system closely matched that of the Digieye color measurement system, suggesting that smartphone-captured images, when processed through this system, can approximate actual colors. Furthermore, the system was applied to a set of 25 solid-color woven fabrics. The color measurements obtained from these samples were then compared with the results from the Digieye system. The comparisons revealed a significant degree of consistency, as measured by the ΔE₁₉₇₆ metric and ΔE₀₀ metric, demonstrating the system's efficacy in fabric color measurement,, demonstrating the system’s efficacy in fabric color measurement. To explore the generalization capabilities of the system across various devices, same-color measurement exercises were conducted using smartphones from different brands, including multi-color fabric samples and the aforementioned solid-color woven fabrics. The color differences between the measurements taken from these different devices and those obtained from the Datacolor650 were analyzed. The results of this analysis consistently showed the adaptability of the system in processing images from different smartphone brands, and its ability to provide accurate and reliable color measurements, were irrespective of the device used.

Conclusion The system was applied for color measurement of woven fabric images, including those captured by smartphones from various brands, encompassing both solid and multi-colored fabrics. The results demonstrated that the color measurements from this system align closely with the Digieye system. For model training in this study, Datacolor Spyde Checkr24 color swatches were used. Compared to the dedicated calibrated color cards of the Digieye system, the number of training samples in this study was limited, impacting the measurement accuracy. Future research will focus on expanding the training sample pool and enhancing the color measurement methodology to increase accuracy and extend the practical application of the system in color-related fields.

Key words: textile fabric, polynomial regression, color measurement, smartphone, clustering algorithm