A technique in Image processing

[a_tek7]

Hi. I know that there's a technique for counting object but I don't how it is.It's based on finding central gravity of objects for counting them.Can any body how this techniques work or its' name?
Sorry for my bad english.

 

[sekonder85]

As I remember this technique based on component labeling. You can count the components at the image by using it.Good luck

 

[jo1776]

The general technique for counting objects in image processing is (assuming you are talking about analyzing a digital image)

1. select a threshold value for the image (a value between the highest pixel value and the lowest pixel value, say the average between the two).

2. set every pixel that is above the threshold value to binary "1".

3. set every pixel below the threshold value to binary "0". You now have a simple binary image.

4. track the perimeter of each object and put it in a little database with appropriate labels etc (starting coordinate, length of perimeter, area, centriod, location coords in image, etc).

the general method for tracking the perimeter is called "chain code". Use a 9 pixel template. Your current location is always the center pixel. Locate a boundary via sequential search for a pixel value not = 0. Placing your center pixel template on the boundary. Search the surrounding 8 pixels to find which direction the boundary continues in. Record this direction. Move your 9 pixel template making this new direction pixel the center of your template. Repeat until you end up back at the point you started.

As you record the chain code perimeter for each object remove those image coordinates from the search space.

The number of objects in you little database equals the number of objects in the image. You can compute centriod or any other geometric property from chain code because no information has been lost.

Before you threshold the image you may want to run a low pass filter (software)over it to smooth out some of the noise.

-jo1776