Image Processing_Boarder Detection （R Studio）
An image is essentially a N*M*3 matrix (like 1024*768*3)
By calculating the local variance of a pixel, and by defining thresholds, it's easy to run a simple boarder detection algorithm.
Note: It works perfectly with anything less furry (check the red-panda and desktop example and how cute little fur-balls could mess up my results.)
Output:
Code:
# Load image
matfile <- readMat("/Users/gaochenying1/Desktop/My Projects Demo/linkedin.mat")
testimage = matfile$maskedRGBImage
dim(testimage)
# Let's see how it goes for a single image
nr=dim(testimage)[1]
nc=dim(testimage)[2]
band=5
N=nr*nc
flag.mat=matrix(0,nrow = (nr-(band-1)),ncol = (nc-(band-1)))
var.mat=matrix(0,nrow = (nr-(band-1)),ncol = (nc-(band-1)))
# Calculate Variance of Neighbours
for (i in 3:(nr-(band-1)/2))
{
for (j in 3:(nc-(band-1)/2))
{
tempmat1=testimage[(i-2):(i+2),(j-2):(j+2),1]
tempvar1=sum((tempmat1-mean(tempmat1))^2)/(band^2)
# tempflag=ifelse(tempvar>)
var.mat[i-(band-1)/2,j-(band-1)/2]=tempvar1
}
}
hist(as.numeric(var.mat), col = "#81D8D0") # Adjust Threshold according to this
# Flag the large variance element and call them borders
thershold=quantile(var.mat,0.92)
for (i in 1:(nr-(band-1)))
{
for (j in 1:(nc-(band-1)))
{
tempflag=ifelse(var.mat[i,j]>thershold,1,0)
flag.mat[i,j]=tempflag
}
}
flag.rt = t(apply(flag.mat, 2, rev))
image(flag.rt,col=c("black", "#9de3e1"), main = "LinkedIn Logo")
remove(list=c("testimage", "matfile", "flag.mat", "var.mat"))
