Quantitative color profiling of digital images with earth mover’s distance using the R package colordistance

Abstract

Biological color may be adaptive or incidental, seasonal or permanent, species- or population-specific, or modified for breeding, defense or camouflage. Although color is a hugely informative aspect of biology, quantitative color comparisons are notoriously difficult. Color comparison is limited by categorization methods, with available tools requiring either subjective classifications, or expensive equipment, software, and expertise. We present an R package for processing images of organisms (or other objects) in order to quantify color profiles, gather color trait data, and compare color palettes on the basis of color similarity and amount. The package treats image pixels as 3D coordinates in a “color space,” producing a multidimensional color histogram for each image. Pairwise distances between histograms are computed using earth mover’s distance, a technique borrowed from computer vision, that compares histograms using transportation costs. Users choose a color space, parameters for generating color histograms, and a pairwise comparison method to produce a color distance matrix for a set of images. The package is intended as a more rigorous alternative to subjective, manual digital image analyses, not as a replacement for more advanced techniques that rely on detailed spectrophotometry methods unavailable to many users. Here, we outline the basic functions of colordistance, provide guidelines for the available color spaces and quantification methods, and compare this toolkit with other available methods. The tools presented for quantitative color analysis may be applied to a broad range of questions in biology and other disciplines.

The fecal microbiota of wild and captive raptors

“…S 1 ), the Earth Mover Distance (EMD) was used as an alternative measure of β-diversity [ 41 ] to avoid distance saturation problems [ 40 ]. EMD values between 1035 pairs (46 × 45/2) of samples were calculated in R with the package Colordistance [ 42 ]. Using graphing program Sigmaplot (Systat Software, San Jose, California), OTU profiles for each sample were converted into histograms such that each OTU was assigned a different color.…”

Genetic and phenotypic evidence of a contact zone between divergent colour morphs of the iconic red‐eyed treefrog

“…We first imported photographs of each frog into Adobe Photoshop CC v. 18.0.0 (San Jose, CA) to correct for ambient light by reference to a black‐white‐grey standard placed in each photo (QPcard 101) and cropped images to only include the posterior surface of the thigh where colour pattern varies (Robertson & Robertson, 2008). Images were then loaded into R (v. 3.5.2) using package colordistance (Weller & Westneat, 2019), which treats images as a set of three‐dimensional points, where each pixel is a coordinate in RGB (red, green, and blue) colour space. Pixel coordinates were binned, and the resulting data included an RGB triplet and percent of pixels for each of eight bins for each individual.…”

Updating salamander datasets with phenotypic and stomach content information for two mainland Speleomantes

“…We suggest the use of the program ImageJ to extrapolate salamanders morphometrics and to estimate their snout-vent length 35 , a fundamental parameter to distinguish juveniles from adults 1 . The images can be also used in R environment ( http://www.R-project.org/ ) to perform analysis on multiple phenotypic traits 51 , 52 . Considering that the dorsal pattern of Speleomantes does not change through the time 33 , it can be used as natural mark to individually recognize salamanders 48 , 53 .…”