Efficiency of drawing rectangles on image matrix in R

问题

I have the function drawRect which will draw a rectangle on a n x m x 3 matrix (one layer for each color channel).

It takes in 2 main parameters: the rectangle params c(xleft, xright, ytop, ybottom) and the image matrix im

drawRect <- function(rect, im, color=2){
  int = 255
  im[rect[3]:rect[4],rect[1],color] = int
  im[rect[3]:rect[4],rect[2],color] = int
  im[rect[3],rect[1]:rect[2],color] = int
  im[rect[4],rect[1]:rect[2],color] = int
  return(im)
}

The function works as its supposed to. However, I am trying to draw ~2000 rectangles on a 3400 x 5200 x 3 image and this is where it becomes EXTREMELY slow.

I have a 2000 x 4 matrix of rectangle parameters which looks something like:

#xleft xright ytop ybottom
313    413  143     243
413    513  143     243
513    613  143     243
613    713  143     243
713    813  143     243
811    911  143     243
...

Any ideas on how to speed this up?...

Note my images are read in using the readJPEG function of the jpeg package and are written to a file using the writeJPEG function.

Edit: I've tried passing in the matrix of rectangle params and using an apply function to avoid multiple calls of the function, but still no significant improvement.

drawRect2 <- function(rects, im, color=2, int = 255){

  x=apply(rects, 1, function(rect){
      im[rect[3]:rect[4],rect[1],color] = int
      im[rect[3]:rect[4],rect[2],color] = int
      im[rect[3],rect[1]:rect[2],color] = int
      im[rect[4],rect[1]:rect[2],color] = int
  })

  return(im)
}

回答1:

I don't know if there is a vectorized version of rect but you can use poylygon which is vectorized . You need just to add NA between rectangles. This answer is inspired from the excellent answer here.

cuts <- function(x)
{
  n <- length(x) %/% 4
  map <- rep(c(rep(TRUE,4),FALSE), n)
  result <- rep(NA, n*5)
  result[map] <- x
  result
}


n <- 2000

xbottom <- runif(n)
ybottom <- runif(n)
xtop <- xbottom+runif(n)
ytop <- ybottom+runif(n)

x <- cbind(xbottom,xbottom,xtop,xtop)
y <- cbind(ybottom,ytop,ytop,ybottom)
cols <- heat.colors(n)[order(x[,1])]
plot(0, xlim=c(min(x),max(x)), ylim=c(min(y),max(y)))
polygon(x=cuts(t(x)), y=cuts(t(y)), col=cols)

This will creates the 2000 rectangles instantaneously.

来源：https://stackoverflow.com/questions/15627674/efficiency-of-drawing-rectangles-on-image-matrix-in-r

标签

image-processing

matrix

rectangles