block matching image issue

When I did blockmatching algorithm on the tennis man image, and did a comparaison between the original and reconstructed image I got this black boxes, why?

[Image Analyst]

There is no way to tell with the minimal information you have shared. If you have any more questions, then attach your image(s) and code to read it in with the paperclip icon after you read this:

TUTORIAL: How to ask a question (on Answers) and get a fast answer

[Blob]

clc

 clear all
close all
load('f1.mat');
load('f2.mat');

figure;
subplot 121
imagesc(F1);
title('image 1');
subplot 122
imagesc(F2);
title('image 2');
colormap gray

 load('e1.mat');
load('e2.mat');

 figure;
subplot 121
imagesc(e1);
title('image 1');
subplot 122
imagesc(e2);
title('image 2');
colormap gray

 [M N]=size(F1);
T=16;
nbr_l=M/T;
nbr_c=N/T;
indbloc_x=1:nbr_l;
indbloc_y=1:nbr_c;
L=length(indbloc_x);
Q=length(indbloc_y);
F2_recons = zeros(size(F2));
for i=1:L
    for j=1:Q
        X=indbloc_x(i);
        Y=indbloc_y(j);
        if i==L
            1+1
        end
        %for n=1:T
        %         bloc2=F2(X:X+T-1,Y:Y+T-1);
        bloc2=F2((1:16)+(i-1)*16,(1:16)+(j-1)*16);
        eqm=inf(15, 15);
        mineqm = inf;
        for i1=-7:7
            for i2=-7:7
                condition = (X+i1) >=1 & (Y+i2) >=1;
                condition2 = ((16+(i-1)*16+i1)<=M) & ((16+(j-1)*16+i2)<=N);
                if ~(condition&&condition2)
                    continue
                else

                    bloc1=F1((1:16)+(i-1)*16+i1,(1:16)+(j-1)*16+i2);
                    eqm(i1+8, i2+8)=(1/(nbr_l*nbr_l))*(sum(sum(bloc2-bloc1).^2));
                    if eqm(i1+8, i2+8)< mineqm
                        %stocke bloc1 et les indices qui correspondent
                        mineqm = eqm(i1+8, i2+8);

                        indx=i1;
                        indy=i2;
                        bloc1min = bloc1;
                    end
                end
            end
        end
        F2_recons((1:16)+(i-1)*16,(1:16)+(j-1)*16) = bloc1min;
    end
end
figure
imshow(uint8(F2_recons))

[Walter Roberson]

We would need your .mat files for testing.

[Blob]

These are the files @Walter Roberson

[Walter Roberson]

(Just showing the output, not trying to debug the problem)

Untitled5

Answer 1

[Walter Roberson]

Your attached code (but not your posted code) has

imshow (F2_recons - e1)

where both of those arrays are double precision and happen to contain integer values. The difference between the two arrays ranges between -255 and +206 . You are passing that array of values -255 to +206 to imshow() as a double precision array. By default, when passed a double precision array, imshow() assumes that the valuable part of the data is the range 0 to 1, and that anything below 0 should be treated as 0 and anything above 1 should be treated as 1. So you more or less end up binarizing the difference array.

You should be considering using

imagesc(F2_recons - e1);
colormap(gray)

[Blob]

@Walter Roberson thank you, I have one more request, could you please show me how to plot the vectors of motion?

[Walter Roberson]

Sorry, no, I do not see anything in your code that corresponds to vectors of motion, or anything that corresponds to motion at all. I cannot tell what you are calculating at all.