Flexible color segmentation of biological images with the R package recolorize

Abstract

Color patterns are easy to observe, but difficult to quantify. A necessary first step for most digital color pattern analysis methods is generating color maps, where every pixel in an image is assigned to one of a set of discrete color pattern elements (color segmentation). Most automatic methods for color segmentation rely on color value thresholds or k-means clustering. While these work in specialized cases, these methods often fail due to biological variation in pattern distribution and intensity, technical variation in images, and poor scalability for batch clustering. As a result, color segmentation is the bottleneck step in a wide variety of analyses. Here we present recolorize, an R package for color segmentation that succeeds in many cases where currently used methods fail. Recolorize has three major components: (1) an effective two-part algorithm for color clustering where color distributions are binned and combined according to perceived similarity; (2) a toolkit for making minor manual adjustments to automatic output where needed; and (3) easy export options to a variety of methods that require accurate color maps. This paper illustrates how to use recolorize, includes examples where we effectively address many of the problems that fail using k-means clustering , and combines the package with two of the most popular R packages for quantifying color pattern variation (patternize and pavo).

Sizing mudsnails: Applying superpixels to scale growth detection under ocean warming

“…But in doing so, unique colour information is lost as an analysable trait. Although we currently do not extract colour categorizations of superpixels, recolorizations might be possible for a broad set of images and support wider trait-based analysis in biology (Weller et al, 2022). Many different algorithmic approaches are possible to segment and extract features from mudsnails or other infauna.…”

Optimum ratio of dietary protein and carbohydrate that maximises lifespan is shared among related insect species

“…Data extraction and analyses were conducted in R version 4.1.3 (R Core Team, 2022) (see Figure 1b). The semi‐automated algorithm to extract simulated data to reconstruct the published lifespan performance landscapes worked as following: The algorithm segmented the performance landscapes with respect to the z ‐axis using the ‘recolorize’ 0.1.0 package (Weller et al., 2022). In this study, the z ‐axis corresponds to lifespan values. Each z‐layer was manually assigned a value for lifespan based on the contour lines depicted in the image on the published literature.…”

Species richness under a vertebral stripe: integrative taxonomy uncovers three additional species of Pholidobolus lizards (Sauria, Squamata, Gymnophthalmidae) from the north-western Colombian Andes

“…of Chicago). To quantitatively illustrate colour differences among the three males selected as new species’ holotypes, we additionally conducted image segmentation analyses on the R package RECOLORIZE ( Weller et al 2021 ), grouping slightly varying hues into a lower number of categories with the method kmeans, and depicting the corresponding result on the sRGB colour space.…”