Roy. G. Biv: The Color Matching Application for Artists With Limited Pigments

When looking at a finished art piece, it is hard to discern what pigments are used to create a particular color. To aid art conservationists and novice artists in color replication, we developed an application that takes in the RGB values of the desired color and calculates the pigment ratios necessary for replicating that color. From a survey of 139 respondents, a total of 86.3% wish that there was a product that would calculate pigments to mix for a specific color. The user interface of the application is familiar and intuitive; it contains a camera screen that averages the RGB values within a crosshair, a screen displaying the calculated pigment ratio, and a color library in which a color and its associated pigment ratio are saved. The application has a 97.8% RGB scanning repeatability, showing that the RGB input is nearly identical each time a color is scanned. To train a machine learning model, a database of 872 hand-painted acrylic entries was constructed using a limited palette. The final training RMSE for the boosted tree model was 0.036 and the final testing RMSE was 0.141. The median color difference in the pigment values between the replicated color and the original color was 0.0668. This shows that the mixed color is 93.32% similar to the desired color. The application not only successfully extracts RGB values from a scanned image to tell the user the necessary pigment values for recreating a color, but also is unique in its non-spectral approach to subtractive color mixing.