RGB to \(L^{ * } a^{ * } b^{ * }\) Color Prediction Model Based on Color Cards

Color is a critical characteristic indicating the traits of an object and is widely used in computer vision technology. In order to ensure uniformity and device independence of color in image processing, standardized and normative color management are needed. \(L^{*} a^{*} b^{*}\) color space is a device-independent expression of color values because it is uniform and independent of hardware devices. Therefore, in this paper, we compare four prediction models from RGB to \(L^{*} a^{*} b^{*}\) . The experimental platform was built to acquire digital images with Pantone color cards as sample data and extract RGB value. The real \(L^{*} a^{*} b^{*}\) value is measured by colorimeter. And preprocess the original RGB value and \(L^{*} a^{*} b^{*}\) value data, and establish four prediction models: linear, quadratic polynomial, neural network and extreme learning machine, by comparing preprocessed data with real data. Using four evaluation metrics (MAE, MSE, RMSE and R2), and scatter fit plots were used to compare results. The results show that the model based on extreme learning machine has the highest prediction accuracy, with MAE reaching 0.021, 0.017 and 0.010, MSE 0.0006, 0.0005 and 0.0001, RMSE 0.026, 0.021 and 0.031, R2 0.96, 0.97 and 0.96, respectively. Indicates that the extreme learning machine has good potential for application.