How to implement this algorithm with region growing ?

Question

hi,
I would like to ask u to help me in developing code for the region growing segmentation algorithm for digital images.The algorithm perform
1.Model I should use the average of the pixel(for each colorchanel,red-green-blue ) from the color I choose.
2. Strating start by choosing an arbitrary seed pixel via 10X10 region and calculate the average of the region and compare it with neighbouring pixels.
3. growing region is grown from the seed pixel by adding in neighbouring pixels that are similar , increasing the size of the region.
4. stop when the growth of one region stops, then the program should try another direction. the growth is starting on the left and in the same direction as the rotating hands of a clock
the whole process is stoped until the growth of all the direction stoped and choose a largest region with a bounding box. all the pixels in the bounding box is now similar.
I really don't know , how can I do it?
Please suggest the tips to develop or code if u have for this algorithm.
Thanks for ur help

Answer 1

Something you need to consider is what sort of data structure you will use to hold the result of a region that you are growing. One possibility is to keep a second black and white image that is simply a mask of the grown region, i.e. black in the areas that have been included and white in the areas not yet covered.
You also need to determine a threshold for similarity. How close in color space must two pixels be in order that you consider them to belong to the same region. You will need a boolean function that compares two colors and returns true if they are similar.
Finally, if you use a mask bitmap to indicate where you have been, you can proceed in the following sort of manner.
You can sweep across the entire image array, first right to left, then top to bottom, then left to right, then bottom to top, cycling through these 4 directions until nothing changes.
A single sweep will consist of a double nested for loop, one incrementing x and the other incrementing y. Inside the inner loop you are considering a single pixel at (x,y)
So assume that you are sweeping from left to right, i.e. incrementing x
You will be expanding the reagion at (x,y) only if
a) mask(x-1,y) is black (i.e. the previous pixel is in the region.
b) maks(x,y) is white (the current point is not in the region
c) color of image at (x,y) is similar to color of region
When you update the region you will
a) color the mask at (x,y) black
b) add the RGB from the image to TotalR,TotalG,TotalB in region
c) increase point count for the region
The point count for a region and the TotalR etc allow you to compute the average color of the region at any time. You can detect if a single pass changed anything by noticing if the pointCount changed between two passes
This algorithm is not particularly efficient, because you keep sweeping over the entire array, but it is easy to comprehend and easy to code.
Note: the hard part is determining a decent threshold for color similarity. everything else is very straight forward.
Enjoy!