Optimize code for a blob-detector (MATLAB)

[beyondlight]

The blobdetector consist of the following functions:

gaussImage: Filters a image with a gaussian filter
gaussPyramid: Uses gaussImage to filter the same image several times with different std.
diffOfGaussians: Uses gaussPyramid to compute the difference between consecutive filtered images
findDogMaximas: Finds local maximas in a image, by setting it to minus it also finds minimum
dogDetector: Finds extreme points from all layers of filtered images and stores them in a 3xN vector.

findDogmaximas takes about 15 min to run and dogDetector takes over 1 hour!

findDogmaximas has a vector that changes size for every loop iteration since I don't know how many extreme potins that exists in an Image.

Computer specifications:
Lenovo Laptop
i7-3632QM Processor
CPU 2.2 GHz
8 GB RAM
64-bits operativesystem

Code:

function maxima = findDogMaximas(dogLower,dogMiddle,dogHigher)

maxima=[];
nrofmaxima=0;
[M ,N]=size(dogMiddle);
          
count_middle=0;
count_lower=0;
count_higher=0;

          
for j=2:1:M-1

        for k=2:1:N-1                      for x=-1:1:1
                        for z=-1:1:1

                               if dogMiddle(j+x,k+z)<dogMiddle
                                     count_middle=count_middle+1;
                               end

                               if dogLower(j+x,k+z)<dogLower
                                   count_lower=count_lower+1;
                               end

                               if dogHigher(j+x,k+z)<dogHigher
                                    count_higher=count_higher+1;
                               end

                        end
                      end

                     sum=count_middle+count_lower+count_higher;
                                          
                                          
                    if sum<26
                      
                      nrofmaxima=nrofmaxima+1;
                      maxima(1,nrofmaxima)=j;
                      maxima(2,nrofmaxima)=k;
                                              
                    end
                                          
                                          

        end
end
      
      
      
end

function detected = dogDetector(im)

tic
k=2.^(1/3);
stds=zeros(1,9);

    for x=1:1:9

        stds(1,x)=k^(x-1);

    end

stds=1.6.*stds;

DoG=diffOfGaussians(im,stds);
n=length(DoG);
detected=[];

        for i=2:1:n-1

            dogLower=DoG{i-1};
            dogMiddle=DoG{i};
            dogHigher=DoG{i+1};

            maxima=findDogMaximas(dogLower,dogMiddle,dogHigher);
            minima=findDogMaximas(-dogLower,-dogMiddle,-dogHigher);

            dog_stdsi=[maxima minima];
            dog_stdsi(3,:)=stds(i);
            detected=[detected dog_stdsi];
           
        end
toc
end

 

[beyondlight]

My bad!

I wrote the if statemens completely wrong. omg :P

if dogMiddle(j+x,k+z)<dogMiddle
count_middle=count_middle+1;

Im referring to the whole dogMiddle! it should only be one point. That is dogMiddle(j,k).

 

[kreil]

I think the first thing you should do here is to analyze the execution of your code with the profiler. All you need to do is type this into MATLAB first:

profile on

This turns on the code profiler for any subsequent code that you execute.

Next, execute your main function call (dogDetector is the main function here that calls all the others right?) Once MATLAB returns the result (however long that is), type the following into MATLAB to close the profiler and view the results:

profile viewer

Of course, if you want more info, type

help profile

You can click through the results to figure out what is bogging your program down. Once you are able to do this, please come back and post about your findings. It will be much easier to ask for help optimizing a snippet of code in a particular function, than to post several functions in their entirety.