%% Run_Me

%% Clearing
clc;
clear;
close all;

%% Initializing
Img_Dataset = [];
Label_Cell = {};
Label_Arr = [];
Ordered_Label = {};
Dataset_Path = 'dataset';
All_Data_Size = 0;


%% Read dataset
Folders = dir(Dataset_Path);
Folders = Folders(3:end);
for in_Folders =  1: length(Folders)
    Cur_Folder = [Folders(in_Folders).folder '\' Folders(in_Folders).name];
    File_Names = dir([Cur_Folder '\*.bmp']);
    Ordered_Label{in_Folders} = strtok(Folders(in_Folders).name,'_');
    for in_Each_Class = 1 : length(File_Names)
       Cur_Img_Path =  [File_Names(in_Each_Class).folder '\' File_Names(in_Each_Class).name];
       img = imread(Cur_Img_Path);
       Org_Size = size(img);
       img = Segment_Letters(img);
       img = imresize(img,Org_Size);
       All_Data_Size = All_Data_Size + 1 ;
       Img_Dataset(All_Data_Size,:) = img(:);
       Label_Cell{All_Data_Size} =  strtok(Folders(in_Folders).name,'_');
       Label_Arr(All_Data_Size) = in_Folders;
    end
end

%% shuffle data
RR = randperm(All_Data_Size);
Img_Dataset = Img_Dataset(RR,:);
Label_Cell = Label_Cell(RR);
Label_Arr = Label_Arr(RR);


%% Divid data to train and test
Train_Ratio = 75;
Train_Size = floor(All_Data_Size*Train_Ratio / 100);
Train_Dataset = Img_Dataset(1:Train_Size,:);
Test_Dataset = Img_Dataset(Train_Size+1:end,:);
Train_Label_Cell = Label_Cell(1:Train_Size);
Test_Label_Cell = Label_Cell(Train_Size+1:end);
Train_Lebel_Arr = Label_Arr(1:Train_Size);
Test_Lebel_Arr = Label_Arr(Train_Size+1:end);


%% Train Neural Network
trainFcn = 'trainscg';
hiddenLayerSize = [30,30];
net = patternnet(hiddenLayerSize, trainFcn);
net = train(net,Train_Dataset',full(ind2vec(Train_Lebel_Arr)));

%% Test Neural Network
y = net(Test_Dataset');
[val,idx] = max(y);
fprintf('------------------- Network Accuracy ----------------------\n');
fprintf('Network Accuracy = %0.2f %%\n',mean(Test_Lebel_Arr == idx)*100);
fprintf('-----------------------------------------------------------\n');

%% save network
save('Variables','net','Ordered_Label');

%% >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> Functions
function [First_Letter,Last_Part_Image] = Segment_Letters(Input_Image)
    Input_Image = Find_Letters(Input_Image);
    Size_2 = size(Input_Image,2);

    for In_Size = 1 : Size_2
        Col_Sum = sum(Input_Image(:,In_Size));
        if Col_Sum == 0
            Temp1 = Input_Image(:,1:In_Size-1);
            Temp2 = Input_Image(:,In_Size:end);
            First_Letter = Find_Letters(Temp1);
            Last_Part_Image=Find_Letters(Temp2);
           break
        else
            First_Letter = Input_Image;
            Last_Part_Image = [ ];
        end
    end
end

function Output = Find_Letters(Input)
    [Val,Size] = find(Input);
    Output = Input(min(Val):max(Val),min(Size):max(Size));
end   

%% Run_Single_Image

%% Clearing
clc;
clear;
close all;

%% Initialize
Label_Str=[ ];
Final_Seg = {};

%% Load image
load('Variables');
[FileName,Path] = uigetfile('*.*');
Loaded_Image = imread([Path,FileName]);
imagen1 = Loaded_Image;

%% Process image
if size(Loaded_Image,3)==3
    Loaded_Image=rgb2gray(Loaded_Image);
end

if ~islogical(Loaded_Image)
    threshold = graythresh(Loaded_Image);
    Loaded_Image = ~imbinarize(Loaded_Image,threshold);
end

Loaded_Image = bwareaopen(Loaded_Image,15);
Temp_Img = Loaded_Image;


%% Image segmentation and classification

while 1
    [First_Line,Temp_Img]=Find_Line_Images(Temp_Img);
    Letter_Num = 0;
    Temp_First_line_Image = First_Line;              
   
    while 1
       [First_Letter,Temp_First_line_Image] = Segment_Letters(Temp_First_line_Image);
       cur_obj = imresize(First_Letter,[25,20]);
       Letter_Num = Letter_Num + 1;
       
       Img_4_Net = cur_obj(:);
       Net_Output = net(Img_4_Net);
       [~,Net_Output] = max(Net_Output);
       Final_Seg{end+1} = cur_obj;
       Label_Str = [Label_Str Ordered_Label{Net_Output}];
       if isempty(Temp_First_line_Image)
           break;
        end
    end
    if isempty(Temp_Img)
        break
    end
end

%% Show result of segmentation and classification
figure;
numb = length(Final_Seg);
m_row = ceil(sqrt(numb));
m_col  =m_row;
for i = 1 : numb
    subplot(m_row,m_col,i);
    imshow(Final_Seg{i})
    title(Label_Str(i))
end

%% >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> Functions
function [First_Letter,Last_Part_Image] = Segment_Letters(Input_Image)
    Input_Image = Find_Letters(Input_Image);
    Size_2 = size(Input_Image,2);

    for In_Size = 1 : Size_2
        Col_Sum = sum(Input_Image(:,In_Size));
        if Col_Sum == 0
            Temp1 = Input_Image(:,1:In_Size-1);
            Temp2 = Input_Image(:,In_Size:end);
            First_Letter = Find_Letters(Temp1);
            Last_Part_Image=Find_Letters(Temp2);
           break
        else
            First_Letter = Input_Image;
            Last_Part_Image = [ ];
        end
    end
end

function Output = Find_Letters(Input)
    [Val,Size] = find(Input);
    Output = Input(min(Val):max(Val),min(Size):max(Size));
end   


function [First_Line_Image,Last_Part_of_Image] = Find_Line_Images(Input_Img)
    Input_Img = Find_Letters(Input_Img);
    Size_1 = size(Input_Img,1);
    for In_Size = 1 : Size_1
        if sum(Input_Img(In_Size,:))==0
            Temp1 = Input_Img(1:In_Size-1, :);
            Temp2 = Input_Img(In_Size:end, :);
            First_Line_Image = Find_Letters(Temp1);
            Last_Part_of_Image=Find_Letters(Temp2);
            break
        else
            First_Line_Image=Input_Img;
            Last_Part_of_Image=[ ];
        end
    end
end