OCR

Optical character recognition program based on deep learning.

How to use

1. Clone this repository with git clone git@github.com:GabRayz/OCR.git.
2. Compile the project by running the command make in the project folder.
3. Execute the program with ./ocr.

Graphical User Interface

The character recognition program can be used through a GUI. Use ./ocr to launch it. Drag and drop an image in the GUI to process it. The extracted text is shown on the left side. The result can then be saved into a txt file.

Command Line Interface

The CLI offers more functionalities. Type ./ocr help to list available commands. Type ./ocr {command} to see the usage of the command.

• ./ocr write_dataset: Creates a new learning dataset from a directory containing images to be segmented. Each image must contain the following string : 0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ'()-_.,?!:;.
• ./ocr learn: Trains a new neural network from a given dataset. The so created network can be saved.
• ./ocr read: Reads an image using the given neural network. The extracted text is displayed and saved into a txt file.

Dependencies

• SDL2 : SDL2 is used to make the Graphical User Interface.
• Image Magick 7 : Import, resize, rotate images.
• Hunspell 7 : Hunspell is a spell checker and morphological analyzer designed for languages with rich morphology and complex word compounding and character encoding.

Features

• Import images
• Preprocessing :
  • Grayscale
  • Noise Reduction
  • Contrast Enhancement
  • Binarization (Otsu's Method)
  • Rotation Detection (Hough's Transformation)
  • Rotation Correction
• Segmentation :
  • Paragraphs & Lines Detection (X-Y Cut)
  • Characters Detection (Connected Component Labeling CCL)
• Recognition :
  • Neural Network Loading
  • Neural Network Recognition
• Postprocessing :
  • Uppercase removal
  • SpellCheck (Hunspell)
• Display (Graphical User Interface)

Collaborators

• Gabriel Rayzal : github.com/GabRayz
• Tony Heng : github.com/TonyHg
• Nathan Cabasso : github.com/Vardiak