peak-to-valley segmentation

i need help for the vertical segmentation of my binary image by using the peak-to-valley method

notice the valley. I need to segment each character when it meet a valley. I saw many thesis that explain this method but it kinda hart to perform in matlab.

clear all, 
close all
clc

f=imread('C:\Users\Yasmin\Documents\MATLAB\Data0.jpg');
f=imresize(f,[300 NaN]);
g=rgb2gray(f);
org=f;
H = fspecial('unsharp');
g2 = imfilter(g,H,'replicate');
w=[0 1 0; 1 -2 1; 0 1 0];
g3=imfilter(g2,w);
g4=1-im2double(g3);
se1=strel('disk',1);
%gi = imdilate(g4,se1);
%figure,imshow(gi)
ge=imerode(g4,se1);            %%%%morphological image processing
%gdiff=imsubtract(gi,ge);
gdiff=mat2gray(ge);
gdiff=conv2(gdiff,[1 1;1 1]);
A=imadjust(gdiff,[0.5 0.7],[0 1],.1);
%B=logical(gdiff);
    [a1 b1]=size(A);

figure,imshow(A)
[h,w,f]=size(A);
org=A;
A=padarray(A,[0 10]);

%c=bwlabel(A);
%max=(max(c))
%im1=(c==8);
%imshow(im1)

%extraction
final=bwareaopen(A,floor((a1/15)*(b1/15)));  
final(1:floor(.9*a1),1:2)=1;
final(a1:-1:(a1-20),b1:-1:(b1-2))=1;
yyy=template(2);
imshow(final)
Iprops=regionprops(final,'BoundingBox','Image');
hold on
for n=1:size(Iprops,1)
    rectangle('Position',Iprops(n).BoundingBox,'EdgeColor','g','LineWidth',2); 
end
hold off
NR=cat(1,Iprops.BoundingBox);   %%Data storage section
[r ttb]=connn(NR);

if ~isempty(r)

      xlow=floor(min(reshape(ttb(:,1),1,[])));
      xhigh=ceil(max(reshape(ttb(:,1),1,[])));
      xadd=ceil(ttb(size(ttb,1),3));
      ylow=floor(min(reshape(ttb(:,2),1,[])));    %%%%%area selection
      yadd=ceil(max(reshape(ttb(:,4),1,[])));
      final1=A(ylow:(ylow+yadd+(floor(max(reshape(ttb(:,2),1,[])))-ylow)),xlow:(xhigh+xadd));
      [a2 b2]=size(final1);
      final1=bwareaopen(final1,floor((a2/20)*(b2/20)));
      imshow(final1)

      Iprops1=regionprops(final1,'BoundingBox','Image');
      NR3=cat(1,Iprops1.BoundingBox);
      I1={Iprops1.Image};
  end

[h,w,f]=size(final1);
org=final1;
%segmetation

i=padarray(final1,[0 10]);

%segmetation
final=bwareaopen(i,floor((a1/15)*(b1/15)));  
final(1:floor(.9*a1),1:2)=1;
final(a1:-1:(a1-20),b1:-1:(b1-2))=1;
yyy=template(2);
imshow(final)
Iprops=regionprops(final,'BoundingBox','Image');
hold on

for n=1:size(Iprops,1)
    rectangle('Position',Iprops(n).BoundingBox,'EdgeColor','g','LineWidth',2); 
end
hold off
NR=cat(1,Iprops.BoundingBox);   %%Data storage section
[r ttb]=connn(NR);

if ~isempty(r)

      xlow=floor(min(reshape(ttb(:,1),1,[])));
      xhigh=ceil(max(reshape(ttb(:,1),1,[])));
      xadd=ceil(ttb(size(ttb,1),3));
      ylow=floor(min(reshape(ttb(:,2),1,[])));    %%%%%area selection
      yadd=ceil(max(reshape(ttb(:,4),1,[])));
      final1=i(ylow:(ylow+yadd+(floor(max(reshape(ttb(:,2),1,[])))-ylow)),xlow:(xhigh+xadd));
      [a2 b2]=size(final1);
      final1=bwareaopen(final1,floor((a2/20)*(b2/20)));
      imshow(final1)

      Iprops1=regionprops(final1,'BoundingBox','Image');
      NR3=cat(1,Iprops1.BoundingBox);
      I1={Iprops1.Image};
  end

[h,w,f]=size(final1);
A=final1

verticalProjection = sum(i, 1);
subplot(2, 2, 1);imshow(i); 
subplot(2,2,3);
plot(verticalProjection, 'b-');
grid on;

above is my code and i attached the Data0

Answer 1

[Image Analyst]

Why not simply use a template and a lookup table? Just examine each segment of each character. Depending on which segments are dark or not, you can simply look up the number in the look up table. Make sure your images are aligned so that the template overlaps the character segments.