[Exploring Art Data 18 and 19 concern parsing and charting the Graves Art Sales data covering Constable. They will be published later.]
Let’s reproduce the functionality of ImagePlot in R . We’ll do this in several stages. In this post we’ll write code to produce statistical information about collections of image files. In the next post we’ll write code to visualize that information. Then we’ll write command line and graphical user interface code for the visualization. Finally we’ll use the command line code to look at how to perform image analysis distributed over the network.
For the statistical analysis code (image-properties.r), first we’ll need to install and load some libraries:
<code>## source("http://bioconductor.org/biocLite.R")
## biocLite("EBImage")
## You may need to install libmagick for EBImage
library("EBImage")
##install.packages("colorspace")
library("colorspace")
</code>
Then we will need to write code to convert from the computer-friendly RGB colourspace to the human-friendly HSB colourspace.
<code>
## Get the r,g,b colour values for all the pixels in the image as a list
imageRgbs<-function(bitmap){
## Get flat lists of red, green and blue pixel values
red<-imageData(channel(bitmap, "red"))
dim(red)<-NULL
green<-imageData(channel(bitmap, "green"))
dim(green)<-NULL
blue<-imageData(channel(bitmap, "blue"))
dim(blue)<-NULL
## Combine these lists into a table of pixel r,g,b values
data.frame(red=red, green=green,blue=blue)
}
## Convert the RGB data.frame to an RGB objects collection
rgbToHsv<-function(rgbs){
as(RGB(rgbs$red, rgbs$green, rgbs$blue), "HSV")
}
</code>
Next we write the code to produce the statistics for each image. R makes this very easy.
<code>
## Calculate the median values for the HSV coordinates
## The colour returned is not a colour in the image,
## it just contains the median values
medianHsv<-function(hsvcoords){
HSV(median(hsvcoords[,"H"]), median(hsvcoords[,"S"]), median(hsvcoords[,"V"]))
}
## Calculate the minimum and maximum values for the HSV coordinates
## Returns a vector of colours, the first containing low values,
## the second containing high values
## These are not colours that appear in the image, they just contain the values
rangeHsv<-function(hsvcoords){
hrange<-range(hsvcoords[,"H"])
srange<-range(hsvcoords[,"S"])
vrange<-range(hsvcoords[,"V"])
c(min=HSV(hrange[1], srange[1], vrange[1]),
max=HSV(hrange[2], srange[2], vrange[2]))
}
## Calculate the standard deviation for the HSV coordinates
## The colour returned is not a colour in the image,
## it just contains the sd for each value
sdHsv<-function(hsvcoords){
hsd<-sd(hsvcoords[,"H"])
ssd<-sd(hsvcoords[,"S"])
vsd<-sd(hsvcoords[,"V"])
HSV(hsd[1], ssd[1], vsd[1])
}
## A good way of getting the min, max, median and other useful values
summaryHsv<-function(hsvcoords){
list(H=summary(hsvcoords[,"H"]),
S=summary(hsvcoords[,"S"]),
V=summary(hsvcoords[,"V"]))
}
</code>
Now we write the code to output those statistics. This is slightly more complex than the simplest possible way of structuring the code would be in order to make the code more robust if we need to change it later.
<code>
## HSV == HSB
## We use brightness for compatibility
## Some columns and column names are also for compatibility
## Load the file and return a vector of named interesting statistics
fileRow<-function(filename){
cat(filename, sep="\n")
img<-readImage(filename)
rgbs<-imageRgbs(img)
hsvs<-rgbToHsv(rgbs)
hsvcoords<-coords(hsvs)
summaryhsv<-summaryHsv(hsvcoords)
sdhsv<-sdHsv(hsvcoords)
sdhsvcoords<-coords(sdhsv)
## NaN for year for now
c("year"=NaN,
## Get the values manually so that we don't rely on position in case
## that ever changes
"hue_min"=summaryhsv$H[["Min."]],
"hue_1st_qu"=summaryhsv$H[["1st Qu."]],
"hue_median"=summaryhsv$H[["Median"]],
"hue_mean"=summaryhsv$H[["Mean"]],
"hue_3rd_qu"=summaryhsv$H[["3rd Qu."]],
"hue_max"=summaryhsv$H[["Max."]],
"hue_stdev"=sdhsvcoords[,"H"][[1]], ## There must be a better way than this
"saturation_min"=summaryhsv$S[["Min."]],
"saturation_1st_qu"=summaryhsv$S[["1st Qu."]],
"saturation_median"=summaryhsv$S[["Median"]],
"saturation_mean"=summaryhsv$S[["Mean"]],
"saturation_3rd_qu"=summaryhsv$S[["3rd Qu."]],
"saturation_max"=summaryhsv$S[["Max."]],
"saturation_stdev"=sdhsvcoords[,"S"][[1]],
"brightness_min"=summaryhsv$V[["Min."]],
"brightness_1st_qu"=summaryhsv$V[["1st Qu."]],
"brightness_median"=summaryhsv$V[["Median"]],
"brightness_mean"=summaryhsv$V[["Mean"]],
"brightness_3rd_qu"=summaryhsv$V[["3rd Qu."]],
"brightness_max"=summaryhsv$V[["Max."]],
"brightness_stdev"=sdhsvcoords[,"V"][[1]])
}
## Create a frame containing interesting information about the images
filesSummaries<-function(filenames, folder){
cat("Processing: ")
cat(filenames, sep=", ")
filepaths<-sapply(filenames,
function(filename) file.path(folder, filename))
## data.frame columns can be different types, so we add the filenames here
## We don't have the strings as factors as if we paste() them as factors
## they are pasted as numbers (levels)
data.frame(filename=filenames,
imageID=1:length(filenames),
t(sapply(filepaths, fileRow)), stringsAsFactors=FALSE)
}
## Print the fileDetails frame to a tab-separated-values file
## This can easily be loaded back into R
printFilesSummaries<-function(fileDetails, outfile=""){
## Build an array of values for the file images,
## and make it a frame with the filenames as a column
cat(paste(names(fileDetails), collapse="\t"), file=outfile)
cat("\n", file=outfile)
for(row in 1:dim(fileDetails)[1]){
cat(paste(fileDetails[row,], collapse="\t"), file=outfile)
cat("\n", file=outfile)
}
cat("Done.", sep="\n")
}
</code>
This code can be run from an interactive R session. For scripting and distribution it can be convenient to have a command-line interface to the code. So in another file (imgstats) we write a simple command-line interface to the code.
First of all we load the image statistics code
<code>
source("image-properties.r")
</code>
Then we parse the command line arguments and make sure that the user has provided reasonable values to the script, quitting with an advisory message if they have not.
<code>
################################################################################
## Parse the command line
################################################################################
args<-commandArgs(TRUE)
if(length(args) != 1){
stop(paste("usage: imgstats [foldername]"))
}
folder<-args[1]
if(folder == "."){
stop("Please pass the name of the folder to process, not '.' .")
}
</code>
Next we call the code from image-properties.r to process each of the files in the folder the user named as an argument to the script.
<code>
################################################################################
## Process the files
################################################################################
## Add more formats to taste. We need to only load image files though
files<-list.files(folder, pattern="*.(jpg|jpeg|tif|tiff|png|gif|bmp)")
stats<-filesSummaries(files, folder)
</code>
Finally we call the code from image-properties.r to save the data to a file (named after the folder that has been processed by the script).
<code>
################################################################################
## Write the data
################################################################################
outfile<-file(paste(folder, ".txt", sep=""), "w")
printFilesSummaries(stats, outfile)
close(outfile)
</code>
To run the code from the command line we have to set its execute file permission:
<code>
chmod +x imgstat
</code>
And now we can generate statistical data about a folder of images ready to visualize.
Here’s an example of the output from the script:
<tt></tt>filename imageID year hue_min hue_1st_qu hue_median hue_mean hue_3rd_qu hue_max hue_stdev saturation_min saturation_1st_qu saturation_median saturation_mean saturation_3rd_qu saturation_max saturation_stdev brightness_min brightness_1st_qu brightness_median brightness_mean brightness_3rd_qu brightness_max brightness_stdev
1905.5_a_mondrian.jpg 1 NaN 0 51.43 55.29 60.23 65.71 330 16.5393230900438 0 0.232 0.2925 0.3066 0.3537 1 0.123994722897117 0.01176 0.2863 0.4549 0.4818 0.702 1 0.246702823821300
1905.5_b_mondrian.jpg 2 NaN 0 36 40.47 94.86 192 360 90.020405312462 0 0.1333 0.2155 0.1979 0.2581 0.7143 0.0935518755834357 0.04314 0.2 0.6118 0.5523 0.8549 0.9725 0.312092338505255
1905.5_c_mondrian.jpg 3 NaN 0 41.54 264 200.3 330 360 139.634180648185 0 0.141 0.2 0.2078 0.2644 1 0.0978851449446736 0.03137 0.08235 0.1608 0.2296 0.2667 1 0.216606346291597
1905.5_d_mondrian.jpg 4 NaN 0 32 42 76.1 67.5 360 90.4207404627883 0 0.2308 0.306 0.3256 0.4 1 0.139508165859475 0.01569 0.1176 0.2627 0.3912 0.698 1 0.296420604122761
1905.5_e_mondrian.jpg 5 NaN 0 72 124.3 135.3 217.2 345 65.5574296333988 0 0.1646 0.25 0.2706 0.3925 1 0.144880395319209 0.01569 0.2 0.3176 0.4501 0.8314 1 0.304302318390967
1905.5_f_mondrian.jpg 6 NaN 0 48 56.84 85.65 70 360 74.8812889958008 0 0.05 0.1452 0.1937 0.3438 0.8333 0.153920183168041 0.01569 0.2863 0.4353 0.4846 0.7059 0.9725 0.242587145239624
1905.5_g_mondrian.jpg 7 NaN 0 38.46 43.08 42.42 46.32 100 5.77930853730472 0 0.324 0.4706 0.4646 0.5972 1 0.164745255193285 0.01176 0.4196 0.5922 0.6044 0.8588 1 0.243507248302950
1905.5_h_mondrian.jpg 8 NaN 0 40 47.14 54.34 60 360 25.19346977218 0 0.07547 0.1946 0.2287 0.3524 1 0.174754958215903 0.003922 0.2902 0.5882 0.5644 0.8157 0.9961 0.300666257269201