Comments on: Image Compression with the SVD in R

By: John Myles White

John Myles White — Sun, 27 Feb 2011 02:26:42 +0000

Yeah, that’s the right interpretation of ‘i’. Sorry for the confusion. I’ve just edited the example code to keep future visitors from getting confused.

By: Dan Brickley

Dan Brickley — Sat, 26 Feb 2011 22:13:08 +0000

Ah got it, the line with the ‘i’ only really makes sense in the loop (although any of the given values can be substituted…

Great writeup btw, really nice to have this running locally rather than just read about it in papers :)

By: Dan Brickley

Dan Brickley — Sat, 26 Feb 2011 22:05:51 +0000

What’s the ‘i’ in green.matrix.compressed <- u[,1:i] %*% diag(d[1:i]) %*% t(v[,1:i]) ?

I get "Error: object 'i' not found"…

By: Michael Sumner

Michael Sumner — Mon, 21 Dec 2009 00:20:40 +0000

Hi, thanks this is great. I modified your example to use rgdal to read the file, which gives you more options for formats.

library(rgdal)

## this could read the TIFF directly
imdata <- readGDAL("image.ppm")

## as.matrix converts from sp's data.frame form to matrix
red.matrix <- as.matrix(imdata[1])
green.matrix <- as.matrix(imdata[2])
blue.matrix <- as.matrix(imdata[3])

## we need to flip vertically (could transpose with t() if necessary)
green.matrix <- green.matrix[,ncol(green.matrix):1]
green.matrix.svd <- svd(green.matrix)

d <- green.matrix.svd$d
u <- green.matrix.svd$u
v <- green.matrix.svd$v

op <- par(mfrow = c(3,2))
for (i in c(3, 4, 5, 10, 20, 30))
{
green.matrix.compressed <- u[,1:i] %*% diag(d[1:i]) %*% t(v[,1:i])

image(green.matrix.compressed, col = heat.colors(255), main = i)

}