Object based phase correlation tracking method.....

[EDIT: 20110524 09:49 CDT - reformat - WDR]

I have been trying to implement object based phase correlation tracking to track birds in my images....and there are some issues with my code...On running the code I got these errors which are quite clear that they is something wrong with the code I wrote to perform erosion...

% To implement Correlation Tracking 
%Author: Harsha Vardhan Rao Avunoori
%Steps to implement Correlation 1.Load Images 2.Convert into Gray Scale
%3.Perform FFT on images 4.Convolve Images 5.Post Processing stage which
%includes Median Filtering Dilation and Erosion

%Loading Input Image 
a = imread('img_9.jpg');
figure(1)
imshow(a)

%Loading Target Image 
b = imread('img_11.jpg');
figure(2)
imshow(b)

size(a)
size(b)
%RGB2GRAY conversion of input image
i = rgb2gray(a);
[r c] = size(i);
figure(3)
imshow(i)
size(i)
%RGB2GRAY conversion of target image
p=rgb2gray(b);
figure(4)
imshow(p)

%Performing FFT of input image and target image 
img_i = fft2(i);
img_p=fft2(p);

figure(5)
plot(img_i)

figure(6)
plot(img_p)

img_f = zeros(r,c);

%Multiplying the FFT of Input and Target Images
 for i=1:r
     for j= 1:c 
         img_f(i,j) = img_i(i,j).*img_p(i,j);
     end
 end

img_if = zeros(r,c);

%Perfrom inverse FFT on multiplied image 

for i=1:r
    for j=1:c 
        img_if(i,j) = ifft2(img_f(i,j));
    end
end

figure(7)
imshow(img_if)

img_if1 = zeros(r,c);

%Perform Median filtering using medfilt2 for a 3 x 3 neighborhood

%Point to note is medfilt2 is not accepting complex values so I took
%abs(img_if) to make it a real value 

img_if1 = medfilt2(abs(img_if),[3 3]);
figure(8)
imshow(img_if1)

%Perform Erosion using imerode function
%Never tried erosion before so took a simple strel

z = ones(4,4)
SE = strel(z)
img_ero=imerode(img_if1,'SE');
%Something is wrong erosion gives out an error
figure(9)
imshow(img_ero)

%Perform Dilation using imdilate function
h = eye(5)
SE = strel(h)

img_dil=imdilate(img_ero,'SE');
figure(10)
imshow(img_dil)

Errors after executing the code

??? Error using ==> strelcheck at 19
Function imerode expected its second input argument,  SE, to be either numeric or logical.

Error in ==> morphop>ParseInputs at 165
se = strelcheck(se,func_name,'SE',2);

Error in ==> morphop at 14
[A,se,pre_pad,...

Error in ==> imerode at 123
B = morphop(A,se,'erode',mfilename,varargin{:});

Error in ==> Test2 at 81
img_ero=imerode(img_if1,'SE');

What might be the problem ?? Any help would be appreciated...

Answer 1

[Sean de Wolski]

'SE' is a two element string

imdilate(img_ero,SE)

[Harsha Vardhan Rao Avunoori]

Hey Sean thanks for letting me know I have changed SE = strel('line',4,45) where 4 = length and 45 = degrees....On execution I don't get an error now but the image is complete white...I think I should look at the Median filtering code.....

Thanks for helping out....

Answer 2

[Sean de Wolski]

I'm just getting ready to leave work for the day, but here are a few things:

don't call fft2/ifft2 in a for-loop! You're taking the DFT/IDFT at each point, not the matrix as a whole:

FFTR = fft2(img1).*fft2(img2); %don't call the variables i as this is the square root of negative 1 and WILL affect your results
PCM = ifft2(FFTR./(3+abs(FFTR))); %peak correlation matrix
[pk idx] = max(PCM);
[row_pk, col_pk] = ind2sub(size(PCM),idx);

[Harsha Vardhan Rao Avunoori]

Sean how can I plot motion field for this phase correlation method ?

[Sean de Wolski]

doc quiver