Connected Neighbourhood / Regional Maxima - Image Processing

Hi there,

I am not able to understand how imregionalmax works for finding the regional maxima. Can anybody brief me.!! Mainly not clear about the connected neighbourhood.

Thanks for the time.

Answer 1

[Harshit]

Hi Sriram, 8-Connected neighborhood means the pixels in left,right,up,down,left up, left down,up right and down right. It works by finding all the connected component of same intensity say the 3x3 matrix of 22. And then checking if these components have a connectivity with intensity value higher. Connectivity now means connectivity by a single element of the components.

Answer 2

[Sriram]

Check out for this.

Let the below matrix be the input

    10 10 10 10 10  
    10 10 10 30 30 
    10 13 13 10 30
    10 13 10 10 10
    10 10 10 10 10

output from the matlab command - imregional max is

    0 0 0 0 0  
    0 0 0 1 1 
    0 0 0 0 1
    0 0 0 0 0
    0 0 0 0 0

whereas according to MY theoretical understanding the output would be

    0 0 0 0 0      since when element (3,2) which is 13 it will have neighb
    0 0 0 1 1               10  10  10
    0 1 0 0 1               10 <13> 13  which will be replace by 1
    0 1 0 0 0               10  13  10
    0 0 0 0 0      whereas for element (3,3)
                            10  10  30 
                            13 <13> 10  which will be 0 since 30 at(1,3)
                            13  10  10

what went wrong on my understanding?

Answer 3

[Andrei Bobrov]

x=[    10 10 10 10 10  
    10 10 10 30 30 
    10 13 13 10 30
    10 13 10 10 10
    10 10 10 10 10]

imregionalmax(x,4)

[Sriram]

     0     0      0     0     0
     0     0      0     1     1
     0     1    < 1 >   0     1
     0     1      0     0     0
     0     0      0     0     0

Answer 4

[Harshit]

Hi Sriram, It won't be. You have to see the entire connected component. Look at 3rd column and 3rd row 13 it is connected to 2nd row and 4th column 30 in a 8-neighbor way. So 13 is connected to 30 and so it will become zero.

[Harshit]

Please consider entire 30 as one element and 13 as other element even if they are connected by single of their unit they are connected.

[Sriram]

Yeah I understand theoretically, but how to implement this sort of logic/algorithm in C or some other platform. Can you refer to any material where this logic/flow is being explained ?

[Harshit]

You can look at the source of matlab function itself. It is the most optimized code. Else the logic will be for every pixel check its 8 neighbor. If any of them is bigger store it in a lut such that a(13,10)=1 which means intensity 13 and 10 are connected. In the second iteration just traverse the matrix and delete elements on the basis of lut.

[Image Analyst]

In this case, as with most m-files, the m-file for the function in the toolbox is merely a wrapper for a mex file that does the actual heavy lifting. So you can't see what it does internally. And for this case, it doesn't give a citation to a published article.

Answer 5

[Image Analyst]

Regarding your example: No, the 13's won't be a local max. There's a trail of pixels (equal or higher in values) that connect the 13's to a higher number (30), so none of the 13's will be a regional max. With 8 connected the trail can go up, down, left, right, or diagonally.