Smoothing - derivatives and variance - document layout analysis

Dear everyone. I have now a prlem and really need your help.

Now I try to code step by step follow the paper: Paper

And there is my code by matlab

I copy here or link below to download all

The image that i test: image1 (test2.png)_ image2 (test3.png)

(i'm sorry I don't know why I can't upload on this website)

%main------------------------------------------------------------------
close all;
clc;
clear all;
f=imread('test2.png');
f=rgb2gray(f);
f=im2double(f);
f=im2bw(f,0.5);
f=~f;

[a b]=size(f);
%makequadtree(f);
[PH,t1,t2] = findPH(f); % t1 and t2 is the number of 0 begin and end
figure,bar(PH);
 s=floor(0.05*length(PH));% kernel size s ??? I don't know
MH = findMH(PH,s);
figure,bar(MH);
FH = findFH(MH);
figure,bar(FH);
Z1 = FindZ1(FH);
[D1,D]=distanceZ(Z1);
[m,V,P,DP] = varianceV(D);

   %quadtree structure---------------------------------------------------
function bw1 = quadtree(f)

[a b]=size(bw); %kich co ban dau
if(rem(a,2)>0)
    c=ones(1,b);
    bw=[bw;c];
end
[u, b]=size(bw); 
if(rem(b,2)>0)
    d=ones(u,1);
    bw=[bw d];
end
[m n]=size(bw); 

bw1=ones(size(bw)/2);
for(i=2:2:m)
    for(j=2:2:n)
        if (bw(i,j)==1||bw(i-1,j)==1||bw(i,j-1)==1||bw(i-1,j-1)==1)
            bw1(i/2,j/2)=0;
        end

      end

end
%bw1
end

%make quadtree---------------------------------------------------------
function [bw1,m,n] = makequadtree(f)
bw1=quadtree(f);
[m n]=size(bw1);
while (m>=100 && n>=100)%min size is 50 50 
    bw1=quadtree(bw1);
    [m n]=size(bw1);
    figure,imshow(bw1);
end
imwrite(bw1,'top level.jpg');
%figure,boundingbox(bw1);
end

%find PH------------------------------------------------------
function [PH,t1,t2] = findPH(f)
%projection horizontal
[m n]=size(f);
PH=zeros(m,1);
for i=1:m
    for j=1:(n)
        PH(i)=PH(i)+f(i,j);
    end
end
t1=0; % clear the white uper border
while (1)
    if PH(1)==0
        PH = PH(2:length(PH));
    t1=t1+1;
    else break
    end
end
n=length(PH);
t2=0; %clear the white lower border
while (1)
    if PH(n)==0
       t2=t2+1;
        m=length(PH)-1;
        PH = PH(1:m);
       n=m;
    else break
    end
end
end

%finf MH------------------------------------------------
function [MH] = findMH(PH,kernel_size)
[m, ~]=size(PH);
MH=zeros(size(PH));
s=kernel_size;
%s=25;
for y=1:m
    for j=(y-floor(s/2)):1:(y+floor(s/2))
        if (j<m &&j>0)
            MH(y)=(MH(y)+PH(j));
        end
    end
end
MH=floor(MH/s);
end

%find FH--------------------------------------------------------
function [FH] = findFH(MH)
m=length(MH);
FH=[;];
for i=2:m-1
    FH(i)=MH(i)-MH(i-1);
end
FH(1)=MH(1);
FH(m)=-MH(m);
FH=transpose(FH);
%figure, bar(FH);
end

%find Z-------------------------------------------------------
function [Z] = FindZ1(FH)
[m  n] = size(FH);
Z=zeros(size(FH));
for i=1:m-1
    if ((FH(i)<0 && FH(i+1)>=0)||(FH(i)>0 && FH(i+1)<=0))
        Z(i)=i;
    end
end
end

%find distance d-------------------------------------------------
function [D1,D] = distanceZ(f)
[b,~]=size(f);
t=1;
D1=[];
for i=1:b
    if f(i)~=0
        D1(t)=f(i);
        t=t+1;
    end
end
D=[];
m=length(D1);
D(1)=D1(1);
t=2;
for i=1:m-1
    D(t)=D1(i+1)-D1(i);
        t=t+1;
end
D=transpose(D);
end

%find variance-----------------------------------------------------------
function [m,V,P,DP] = varianceV(D)
n = length(D);
m=sum(D)/n;
V1=0;
for i=1:n 
    V1 = V1+((D(i)-m)*(D(i)-m));
end
V=V1/n;
P = 2-(2/(1+exp(-V)));
DP=0;
if P > 0.5
    DP=1;
end
end

http://www.mediafire.com/download/u9x83qe3na95ket/code+matlab.zip

Run with main But I have some problem here:

In this paper, at page 3 (3.2 Periodicity estimation), find MH, I see that s is kernel size but I not clear what is kernel size or kernel size of what ? I think it's very important because when I try with s = 0.05 of length of PH it make something wrong, my variance always big although I use very beautiful image. (test2.png) - (no skew detection needed) I think there is an problem with MH or find FH. I try with a lot of s but not right. If you have any comment please tell me and I can fix this problem

Thank you so much

[Star Strider]

Use the ‘paperclip’ icon above the window in your original post to upload the paper and uncompressed code to the MathWorks website.

[Tran Tuan Anh]

I'm sorry, but there are some function in different file and image in this folder so I don't know how to post it. I will remember and post file by file. Thank you so much

Answer 1

[Tran Tuan Anh]

Please, help me