Raw data set to evaluate a stepwise color correction (CC) pipeline for controlled imaging environments. The workflow integrates flat-field correction (FFC), gamma correction (GC), and white-balance correction (WB), followed by a color-mapping (CM) stage using machine-learning regression—linear, partial least squares (PLS), and neural networks (NN)—to deliver reliable CC in digital images. The pipeline reduces perceptual color differences in the corrected images. An NN with a second-degree polynomial expansion consistently outperformed other CM methods, yielding the lowest color errors and robust performance across varying imaging conditions.

Tests showed that illumination quality and placement are critical: although the common 45° geometry produces favorable uncorrected images, top-mounted area lighting combined with FFC yielded the best corrected color. Imaging-environment materials also mattered; object background color and sidewalls affected fidelity, with diffusely reflective white performing best. Applied to various colored fruit samples, the proposed pipeline produced more consistent fruit colors across illuminants. An open-source Python package (https://github.com/collinswakholi/ColorCorrectionPackage) and an interactive user interface (https://github.com/collinswakholi/ColorCorrectionPackage_UI) implementing this pipeline is available, enabling reproducible analyses and straightforward adaptation to other controlled imaging tasks. Overall, the pipeline improved color reproduction and measurement in digital images and helped bridge the gap between sophisticated CC methods and practical, routine applications.