How to set dynamic threshold for segmentation purpose (MATLAB)

I need to segment 300 hundred word images, contain cursive characters. I've already try a segmentation code, using vertical projection method. It's good for isolated character, but not really work for cursive character.

But, finally we could segment the cursive character based on it's baseline. Using histogram that represent the profile of image, we could define the threshold to segment the image based on its baseline. My recent code, is as follows.

%%Binarization %%
level = graythresh (a);
b = im2bw (a, level);

%%Complement %%
c = imcomplement (b);

%%PadArray %%
i=padarray(c,[0 10]);

%%Vertical Projecttion for Character Segmentation
verticalProjection = sum(i, 1);
set(gcf, 'Name', 'Segmentation Trial', 'NumberTitle', 'Off') 
subplot(2,2,1);imshow(i); 
subplot(2,2,3);
plot(verticalProjection, 'b-');
grid on;

% *Defining the threshold to determine baseline area* % 
threshold=max(verticalProjection)/3;
% threshold = 0; for isolated characters
thresholdedProjection=verticalProjection > threshold;
count=0;
startingColumns = [];
startingColumnsIndex=0;
for j =1:length(thresholdedProjection)
    if thresholdedProjection(j)
        if(count>0)
            startingColumnsIndex=startingColumnsIndex+1;
            startingColumns(startingColumnsIndex)= j-floor(count/2);
            count=0;
        end
    else
        count=count+1;
    end
end
endingColumns=[startingColumns(2:end)-1 j-floor(count/2)];

y=1;
% *Extract each region, result of segmentation process* 
for k = 1 : length(startingColumns)

      % Get sub image of just one character
      subImage = i(:, startingColumns(k):endingColumns(k)); 
      % im = subImage;
      s = subImage;

      % Normalization using algorithm 2 %
      p = normalization2 (s);

      subplot(2,2,2); 
      imagesc (p); 
      axis equal off;
      pause (1); 
      %   figure, 
      imshow (p);

      % Morphological Operation - Thinning %
      t = bwmorph(p,'thin',Inf);  
      n =1;

Sample images:

My problem is, the threshold of all image are vary, different each other (although there are several have the same threshold). That's why I think, I need a dynamic threshold, but what I've tried, still didn't work. Any help and suggestion would be very appreciated.

Thank you so much.

Answer 1

[Image Analyst]

Just copy what someone else already invented and published. No need for you to reinvent the wheel. Find their algorithms here: http://www.visionbib.com/bibliography/contentschar.html#OCR,%20Document%20Analysis%20and%20Character%20Recognition%20Systems