 |
Author |
Topic: Connected components (Read 839 times) |
|
gabon
|
Connected components
« on: Nov 5th, 2004, 9:06am » |
|
I don't know if it is the right forum, anyway, I'm experimenting a little bit on computer-vision and I'm trying to learn the notions about image processing. Now I'm trying to use the connected component algorithm http://homepages.inf.ed.ac.uk/rbf/CVonline/LOCAL_COPIES/OWENS/LECT3/node 2.html#SECTION00021000000000000000
but obviously I can't make it work.
Instead to use just flags I went straight using colors, so the conceptual calculate is:
Code:
if(A == black){
do nothing;
}else if(D != black){
A = D;
}else if(B == black && C == black){
color++;
A = color;
// Here an issue
}else if(B != black || C != black){
A = color;
// I tried also A = B; or A = C;
}else if(B != black && C != black){
if(B = C){
A = B;
}else{
B = C = A;
}
}
|
|
Does someone see the mistake?
Thank you very much, chr
|
|
|
|
|
zach
|
Re: Connected components
« Reply #1 on: Nov 5th, 2004, 12:48pm » |
|
I'm not sure about errors, but my students just did connected components in my computer vision class. they recreated the algorithm described quite well by David Eberly inside of this code:
http://www.magic-software.com/Source/ImageAnalysis/WmlBinary2D.cpp
It works by labeling groups across the horizontal, then linking the horizontal groups together in a very clever way.
maybe this can help?
- zach
|
|
|
|
|
gabon
|
Re: Connected components
« Reply #2 on: Nov 5th, 2004, 1:23pm » |
|
Also mine works in scrolling pixels[], we can say. The code of your students seems little bit more complex  either to read and to apply. There is a big difference between P5 and C, so I would like to keep it lighter as possible.
Thank you very much anyway 
chr
|
|
|
|
|
zach
|
Re: Connected components
« Reply #3 on: Nov 5th, 2004, 10:41pm » |
|
actually, the code has a straightforward description in the comments which could be pretty easily implemented in java...
|
|
|
|
|
gabon
|
Re: Connected components
« Reply #4 on: Nov 7th, 2004, 12:52pm » |
|
I found an other algorithm, that works better but not perfectly. So starting from a b/w image at 1 bit I've implemented in this way:
Code:
labels = new int[npixels];
for(int i=0; i<npixels; i++){
if(pixels[i]==black){
labels[i]=0;
}else{
labels[i]=1;
}
}
//
int NewLabel=0;
int row,rowt,rowb;
int lp,lq;
int lx=1;
for(int y=1; y<h; y++){
row=y*w;
rowt=(y-1)*w;
for(int x=1; x<w; x++){
if(pixels[row+x]==white){
lp=labels[rowt+x];
lq=labels[row+(x-1)];
if(lp == 0 && lq == 0) {
NewLabel++;
lx = NewLabel;
}else if((lp != lq)&&(lp != 0)&&(lq != 0)){
lx = lq;
}else if(lq != 0){
lx = lq;
}else if(lp != 0){
lx = lp;
}
labels[row+x] = lx;
}
}
}
println(NewLabel);
for(int i=0; i<npixels; i++){
pixels[i]=color(labels[i],labels[i],labels[i]);
}
|
|
very strange... thx, chr
|
|
|
|
|
|
