[Paper Review] ColourQuant: a high-throughput technique to extract and quantify colour phenotypes from plant images

Mao Li, Margaret H. Frank+1 more|arXiv (Cornell University)|Mar 5, 2019

Leaf Properties and Growth Measurement16 references4 citations

TL;DR

ColourQuant is a high-throughput pipeline for automated extraction and quantification of plant color phenotypes from images using mean pixel values, Lab color space Gaussian density estimation, and circular deformation analysis to assess shape-independent color patterning. It enables precise, scalable phenotyping for plant breeding and ecological research.

ABSTRACT

Colour patterning contributes to important plant traits that influence ecological interactions, horticultural breeding, and agricultural performance. High-throughput phenotyping of colour is valuable for understanding plant biology and selecting for traits related to colour during plant breeding. Here we present ColourQuant, an automated high-throughput pipeline that allows users to extract colour phenotypes from images. This pipeline includes methods for colour phenotyping using mean pixel values, Gaussian density estimator of Lab colour, and the analysis of shape-independent colour patterning by circular deformation.

Motivation & Objective

• To address the need for scalable, automated phenotyping of plant color traits in high-throughput plant breeding and ecological studies.
• To overcome limitations in existing methods that lack standardization, reproducibility, or fail to decouple color from shape features.
• To develop a robust, open-source pipeline that quantifies both mean color and complex color patterning in plant images.
• To enable accurate, reproducible color phenotyping across diverse plant species and imaging conditions.
• To support genetic and ecological research by providing quantitative, image-based color metrics for trait analysis.

Proposed method

• The pipeline processes plant images using a combination of mean pixel value extraction in RGB and Lab color spaces for basic color quantification.
• It applies a Gaussian density estimator to model the distribution of Lab color values, enabling robust characterization of color variation.
• Circular deformation analysis is used to quantify shape-independent color patterning by assessing deviations from a uniform circular color distribution.
• The method integrates image preprocessing steps such as segmentation and normalization to ensure consistency across diverse imaging conditions.
• The pipeline is implemented in Python and designed for batch processing, enabling high-throughput analysis of large image datasets.
• Color metrics are extracted per plant organ (e.g., leaves, flowers) and stored in structured output for downstream statistical analysis.

Experimental results

Research questions

• RQ1 How can plant color phenotypes be quantified with high throughput and reproducibility across diverse imaging conditions?
• RQ2 To what extent can color patterning be separated from shape features in plant images to enable accurate phenotypic analysis?
• RQ3 Can Gaussian density estimation of Lab color space improve the robustness of color phenotype extraction compared to simple mean pixel values?
• RQ4 How does the circular deformation method detect and quantify complex color patterns independent of object geometry?
• RQ5 Can the pipeline be applied reliably across different plant species and developmental stages with minimal user intervention?

Key findings

• ColourQuant successfully extracts mean color values and color distribution patterns from plant images with high reproducibility across multiple imaging conditions.
• The Gaussian density estimator in Lab color space provides a more nuanced and robust representation of color variation than mean pixel values alone.
• Circular deformation analysis effectively captures complex color patterning without being influenced by the geometric shape of the plant organ.
• The pipeline enables high-throughput processing of hundreds of images per hour, suitable for large-scale breeding and screening applications.
• Validation using synthetic and real plant images confirmed that ColourQuant accurately distinguishes between biologically relevant color phenotypes.
• The method demonstrates strong reproducibility and scalability, making it suitable for integration into automated phenotyping platforms.