## page was renamed from Matlab
{{{
%melog.m

% ME/MC using four-step log search 



clear all;

close all;

load bus_frames frame;



figure(1);

frame1=frame(:,:,1);

imagesc(frame1, [0,255]);

colormap(gray);



figure(2);

frame2=frame(:,:,2);

imagesc(frame2, [0,255]);

colormap(gray);







temp=size(frame);

dimy=temp(1);   %height of frame

dimx=temp(2);   %width of frame



blockx=16;      %block sizes

blocky=16;



srx=15;         %size of search window +-

sry=15;



%intialize mvf matrix

matchy=zeros(dimy/blocky,dimx/blockx);

matchx=zeros(dimy/blocky,dimx/blockx);



tic;

for loopi=2:dimy/blocky-1,

	%loopi

	for loopj=2:dimx/blockx-1,

		ybound1=(loopi-1)*blocky+1;

                ybound2=loopi*blocky;

                xbound1=(loopj-1)*blockx+1;

                xbound2=loopj*blockx;



                %divide frame2 into blocks

                block=frame(ybound1:ybound2,xbound1:xbound2,2); %current block



		currentoffy=0; currentoffx=0;	least_mse=inf;



		for steps=[2 4 8 16],	% four step search

		for offoffy=-ceil(sry/steps):ceil(sry/steps):ceil(sry/steps),	%offset vector

		    for offoffx=-ceil(srx/steps):ceil(srx/steps):ceil(srx/steps),	

			offy=currentoffy+offoffy;

			offx=currentoffx+offoffx;

			if (ybound1+offy>=1&ybound2+offy<=dimy...

                                &xbound1+offx>=1&xbound2+offx<=dimx)

				prev_block=frame(ybound1+offy:ybound2+offy,xbound1+offx:xbound2+offx,1);

				current_mse=sum((prev_block(:)-block(:)).^2);

				if (current_mse<least_mse)	%found currently least mse

					least_mse=current_mse;

					bestoffy=offy; 

					bestoffx=offx;

				end

			end

		    end

		end

		currentoffy=bestoffy;

		currentoffx=bestoffx;

		end

    	matchy(loopi,loopj)=-bestoffy;

        matchx(loopi,loopj)=-bestoffx;



	end

end	



time_log=toc;



figure(6);

quiver(matchx,matchy);

axis ij;axis image;

title('Motion Vector Field, BM with Four-step Search');





%MC prediction

predict=zeros(dimy,dimx);



for loopi=1:dimy/blocky,

        %loopi

        for loopj=1:dimx/blockx,



                ybound1=(loopi-1)*blocky+1;

                ybound2=loopi*blocky;

                xbound1=(loopj-1)*blockx+1;

                xbound2=loopj*blockx;



                offy=-matchy(loopi,loopj);

                offx=-matchx(loopi,loopj);



                predict(ybound1:ybound2,xbound1:xbound2)...

		=frame(ybound1+offy:ybound2+offy,xbound1+offx:xbound2+offx,1);

        end

end



figure(3);

%colormap(gray(256));

%subplot(2,1,1);

%imagesc(frame(blocky+1:dimy-blocky,blockx+1:dimx-blockx,2));

%axis image;

%title('Current Frame');



%entropy of motion vector fields



dy=matchy(2:dimy/blocky-1,2:dimx/blockx-1);

dx=matchx(2:dimy/blocky-1,2:dimx/blockx-1);

rangey=min(dy(:)):max(dy(:));

rangex=min(dy(:)):max(dy(:));

[county,tmp]=hist(dy(:),rangey);

[countx,tmp]=hist(dx(:),rangex);

proby=county/sum(county);

probx=countx/sum(countx);

proby(proby==0)=1;

probx(probx==0)=1;

H=-sum(proby.*log(proby)/log(2))-sum(probx.*log(probx)/log(2))







%PSNR between the two frames

DFD=abs(frame(:,:,2)-frame(:,:,1));

DFD_ins=DFD(blocky+1:dimy-blocky,blockx+1:dimx-blockx);

psnr1=sum(sum(DFD_ins.^2));

PSNR_frame=10*log10(255*255*(dimy-2*blocky)*(dimx-2*blockx)/psnr1)



DFD=abs(frame(:,:,2)-predict);

%inside part of the error image

DFD_ins=DFD(blocky+1:dimy-blocky,blockx+1:dimx-blockx);

psnr1=sum(sum(DFD_ins.^2));

PSNR_log=10*log10(255*255*(dimy-2*blocky)*(dimx-2*blockx)/psnr1)



%subplot(2,1,2);

imagesc(predict(blocky+1:dimy-blocky,blockx+1:dimx-blockx));

colormap(gray(256));

axis image;

str=sprintf('Prediction from Prev Frame and ME\nBM Method, Four Step Log Search, PSNR=%5.2fdB',PSNR_log);

title(str);



time_log



%save bus_data time_log PSNR_log -append; 

%===================================================================



}}}