A quick and dirty method that may not always work is this:
IM=imread('li.jpg');
IM=IM(321:472,97:1106,:);%crop white field
IM_BW=mean(IM,3)/255;%convert to grey scale double
IM_projection=logical(max(IM_BW,[],1));
idx_end=find(diff(IM_projection)<0);%col index where a letter ends
idx_start=find(diff(IM_projection)>0);%col index where a letter ends
%This may not be true if the letters are kerned
distance=idx_start(2:end)-idx_end(1:(end-1));
split_col_idx=idx_end([distance>50 false])+1;%50 is a threshold distance
%split_col_idx contains the column where you should split the words
The result here is that you should split at columns 465 and 713 in the cropped image, which is indeed after the y and the s.
