Step 2 - Motion Vector Estimation

Lets consider the example below

Now lets test the above scenario with our code.

function [motion_vec,y_pos,x_pos] = motion_estimate(image1,window, block,block_index)

%input the macroblock from image2 to be searched and the window[x0 y0 x1 y1].As results it provide the motion vector in [y x] format

motion_vec=[];

score=10000000000000000;

y_pos=0;x_pos=0;

for y=window(2):window(4)

for x=window(1):window(3)

mmse = sum(sum(abs(cell2mat(block)-cell2mat(image1(y,x))).^2));

if(mmse<score)

score=mmse;

motion_vec = [(block_index(1)-y) (block_index(2)-x)];

y_pos = y;

x_pos = x;

end

end

end

end

%mse = sum(sum(abs(h.^2-k.^2)))

It works! Now we have to find a way to do this to the entire image by moving the window appropriately.

Step 3 - Motion Vector Array

[IMG1,img_y,img_x] = macroblocks(frame1,macroblock_size);

[IMG2,img_y,img_x] = macroblocks(frame2,macroblock_size);

mv_array = MotionVecArray(IMG1,IMG2,4);

Now lets try the simple example we considered above.

function mv_array = MotionVecArray(image1,image2,w)

%image1 - cell array if image1

%image2 - cell array if image2

%w= window size - 4

mv_array = {};

window = ones(1,4);

for y=1:size(image2,1)

for x=1:size(image2,2)

%-------determining window limits------------------------------

%-------x0 in window-------------------------------------------

if (x-w<1)

window(1) = 1;

else

window(1) = x-w;

end

%-------y0 in window-------------------------------------------

if (y-w<1)

window(2) = 1;

else

window(2) = y-w;

end

%-------x1 in window-------------------------------------------

if (x+w>size(image2,2))

window(3) = size(image2,2);

else

window(3) = x+w;

end

%-------y1 in window-------------------------------------------

if (y+w>size(image2,1))

window(4) = size(image2,1);

else

window(4) = y+w;

end

%motion vector determination for each block

[motion_vec,~,~] = motion_estimate(image1,window,image2(y,x),[y x]);

mv_array(y,x)={motion_vec};

end

end

end

Step 4 - Motion Compensation

motionCompensatedIMG = motionCompensation(IMG1,mv_array);

Lets consider the example we considered through out. We can see that from motion vector array and previous image which is image 1 it has reconstructed the image 2. Notice that the [5 5;5 5] is missing.

Now lets do the same on a real image.

function motionCompensatedIMG = motionCompensation(image1,mvarray)

motionCompensatedIMG={};

for y=1:size(image1,1)

for x=1:size(image1,2)

temp=cell2mat(mvarray(y,x));

y_pos = y+temp(1);

x_pos = x+temp(2);

motionCompensatedIMG(y,x)=image1(y_pos,x_pos);

end

end

end