OCR Word Search Solver

Optical Character Recognition (OCR) project developed entirely C.

This software solves Word Search puzzles using image processing and a custom neural network.

Technical Implementation

Core Features

• Language: C
• Neural Network: Multilayer Perceptron (MLP) implemented from scratch (training & inference).
• GUI: GTK+ 3 interface with a step-by-step wizard.
• Image Processing: SDL2 & GdkPixbuf for pixel manipulation.

Detection Pipeline

• 1 - Preprocessing: Grayscale, binarization, and noise reduction.
• 2 - Segmentation:
  • Grid: Blob detection algorithm with median-based filtering.
  • Words: Histogram projection profiles.
• 3 - OCR: 48x48 normalized input fed into the MLP.
• 4 - Solving: Backtracking algorithm to find words in the grid.

Build & Run

Dependencies

• GCC, Make
• GTK+ 3.0
• SDL2, SDL2_image

Usage

cd src
make
./_gui

Project Structure

• src/detect: Segmentation algorithms.
• src/nn: Neural network implementation.
• src/ocr: Main pipeline logic.
• src/solver: Grid solving algorithm.
• src/gui: User interface.

---