Color calibration of digital images for agriculture and other applications

Image processing in agriculture relies primarily on correlating color changes within images’ region of interest to specific quality attributes (e.g., plant phenology, plant health, crop stress, maturity). Changes in lighting conditions during image acquisition affect image color, even though there is no change in quality, and produces misleading inference when used without calibration. The focus of this study was to develop a method for calibrating images to make them homogeneous to improve phenological comparisons. The method was developed with synthetic images and validated with actual plant images in laboratory and field conditions using a standard ColorChecker (X-Rite) chart. Six different color schemes were tested to determine the effect of patch order, and minimum number of patches required for efficient calibration. A user-coded ImageJ plugin named ‘ColorCal’ was developed in Fiji package for color calibration that derived and applied a [3×3] color calibration matrix, based on selected color patches and standard values. Modified total error and calibration performance index (CPI) were developed to evaluate calibration performance. Calibration using any 12 color patches taken in any order gave equal performance (0.14⩽CPI⩽0.26). Calibration performance using only commonly followed neutral color patches (e.g., white, gray) was poor (0.26⩽CPI⩽1.0). Using red (R), green (G), and blue (B) color patches was recommended as it produced visually similar images, the performance was comparable with 24 color patches (0.21⩽CPI⩽0.24), and was simple and practical. The developed plugin took ≈7 s for calibration (Windows laptop, Intel Core i5, and 8 GB RAM). Determining phenological and other applications using the plugin was more reliable than using the raw images.