catsim: a Categorical Image Similarity Index

The goal of catsim is to provide a similarity measure for binary or categorical images in either 2D or 3D similar to the MS-SSIM index for color images. Suppose you have a ground truth segmentation of some image that has been segmented into regions - perhaps a brain scan with different types of tissues or a map with different types of terrain - and a segmentation produced by some classification method. Comparing the two pixel-by pixel (or voxel-by-voxel) might work well, but a method that captures structural similarities might work better for your purposes. MS-SSIM is an image comparison metric that tries to match the assessment of the human visual system by considering structural similarities across multiple scales. CatSIM applies a similar logic in the case of 2-D and 3-D binary and multicategory images, such as might be found in image segmentation or classification problems.

Installation

You can install the released version of catsim from CRAN with:

install.packages("catsim")
#### or the dev version with:
#devtools::install_github("gzt/catsim")

Usage

If you have two images, x and y, the simplest method of comparing them is:

library(catsim)
set.seed(20200505)
x <- besag
y <- x
y[10:20,10:20] <- 1
catsim(x, y, levels = 3)
#> [1] 0

By default, this performs 5 levels of downsampling and uses Cohen’s kappa as the local similarity metric on 11 x 11 windows for a 2-dimensional image and 5 x 5 x 5 windows for a 3-D image. Those can be adjusted using the levels, method, and window arguments.

Please note that the catsim project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.