ColorBayes: Improved color correction of high-throughput plant phenotyping images to account for local illumination differences

Abstract

Abstract Background Color distortion is an inherent problem in image-based phenotyping systems that are illuminated by artificial light. This distortion is problematic when examining plants because it can cause data to be incorrectly interpreted. One of the leading causes of color distortion is the non-uniform spectral and spatial distribution of artificial light. However, color correction algorithms currently used in plant phenotyping assume that a single and uniform illuminant causes color distortion. These algorithms are consequently inadequate to correct the local color distortion caused by multiple illuminants common in plant phenotyping systems, such as fluorescent tubes and LED light arrays. We describe here a color constancy algorithm, ColorBayes, based on Bayesian inference that corrects local color distortions. The algorithm estimates the local illuminants using the Bayes’ rule, the maximum a posteriori, the observed image data, and prior illuminant information. The prior is obtained from light measurements and Macbeth ColorChecker charts located on the scene. Results The ColorBayes algorithm improved the accuracy of plant color on images taken by an indoor plant phenotyping system. Compared with existing approaches, it gave the most accurate metric results when correcting images from a dataset of Arabidopsis thaliana images. The software is available at https://github.com/diloc/Color_correction.git .

Plant phenotyping relevance

植物フェノタイピング画像の局所的な色歪みを補正するアルゴリズムを開発し、既存手法およびArabidopsis画像データセットで精度を検証しているため、方法が中心的である。

Code and data availability

The paper's ColorBayes color-correction algorithm code is explicitly stated as publicly available on the authors' GitHub repository. The green fabric ground-truth image dataset and Arabidopsis plant image datasets are described but no public deposit is stated for them.