This is a module of our project for image processing course at Cairo university.

This is a working in progress project.

The implementation of this tool is from the following paper "A Statistical approach to line segmentation in handwritten documents" Manivannan Arivazhagan, Harish Srinivasan and Sargur Srihari.

It can be found at this link.

This tool converts handwritten text image (including straight and skewed lines) to text. Afterwards, It'll afterwards integrated with a machine learning tool to detect characters.

The main aim of this project is to compile and execute (try to repair compilation errors too! :) ) code handwritten on paper

• OTSU Threshold and Binarization.

• Get initial candidate lines using y-axis histogram projection profile and using adaptive threshold between valleys. These are examples of the output produced from the current code.

• Apply line drawing algorithm and make decissions about regions that hit the lines using Bivariate Gaussian Distribution. These is example of the output produced from the current code.

This is an example of the output as labeled components for MatchScore evaluation.

• OpenCV: 3.0 or newer.

mkdir build
cd build
cmake ..
make
./Image2Lines the_path_to_the_img the_path_to_the_output_directory

the output will be:

• N final lines, each line is named 'line_i.jpg' where i is the line number.
• 'Initial_lines.jpg' : representing the initial lines drawn on the original image.
• 'Final_lines.jpg' : representing the final lines drawn on the original image.
• 'Contours.jpg' : represnting the detected contours.

You should add LineSegmentation.cpp and LineSegmentation.hpp to your project.

    #include "LineSegmentation.hpp"
    
    // Create an object from LineSegmentation and give to it the image path
    LineSegmentation line_segmentation(img_path);
    
    // Run segment() method to segment lines. The output will be a vector of CV::Mat each representing a line.
    vector<cv::Mat> lines = line_segmentation.segment();
    

NOTE You should link correctly OpenCV in your working project. if you use CMake see CMakeLists.txt file to see how to add opencv.

mkdir build
cd build
cmake ..
make

• Finding peaks and valleys can cause in some cases the lines to be falsely segmented. If you can improve that problem, feel free to make a pull request.

• Improve detecting peaks and valleys function.
• Use a better binarization technique than the currrently used (OTSU) in order to get better binarization in case of camera images. (I will use my other repository named Binarization).

IAM database (Educational use permit).