Second Project AI

🚀 Final Project: Image Text Recognition & Sentiment Analysis System

This repository contains the details of the Image Text Recognition and Sentiment Analysis

Contents

🎯 Project Objective**

Build a two-stage AI pipeline that can:

1. Read Text from an Image 📷

   • Leverage a model trained on the IIIT-5K Words dataset to detect and transcribe any text in an image (e.g., a sign, screenshot, or advertisement).

2. Analyze Sentiment ❤️ 😭 😐

   • Feed the extracted text into a Recurrent Neural Network (RNN) trained on the Twitter Sentiment Dataset to classify it as positive 👍, negative 👎, or neutral 😐.

The goal is to create an end-to-end system that can

• “see” an image 👁️
• “read” its contents 📝
• “understand” the tone of the message 🧠

📊 Dataset: IIIT5K-Words

🔗 Source & Download

• Curated by IIIT-Hyderabad (IIIT-H).
• Distributed as a .tar.gz archive with annotations in MATLAB .mat files.
• Official download: IIIT5K-Words official site.

🗂️ Size & Structure

• Total images: 5,000 single‐word crops
• Suggested split:
  • Train: 3,000 images
  • Test: 2,000 images
• Image format: JPG/PNG, each containing one isolated word
• Annotations:
  • testdata.mat → list of image paths + word labels
  • testCharBound.mat → per‐character bounding‐box coordinates

🔣 Content & Complexity

• Typographic variability:
  • Multiple fonts, sizes and styles (italic, bold, serif, sans-serif)
• Real-world challenges:
  • Noisy or semi-transparent backgrounds
  • Partially occluded characters
  • Compression artifacts and blur

🏷️ Labeling

• Each image shows exactly one English word.
• Ground‐truth

❓ The Missing Data Challenge

Although the standard Wine Quality Dataset does not include missing values, in real-world scenarios it is very likely that some physicochemical measurements may be absent when evaluating a wine. Therefore, a fundamental aspect of this application is its capacity to manage the absence of one or more input values provided by the user. 🤷‍♂️

Requirements

• Python
• Dataset OCR: The dataset was provided in the following link OCR Images
• Dasatet Sentiments: The dataset was provided in the following link Sentiment dataset

Installation

1. Clone the project on your computer:

   git clone https://github.com/C102002/proyecto-ia-2

Note

Python Version 3.11 🚀:

• Dependency Compatibility: Using Python 3.11 helps resolve known issues with data analysis and dependency conflicts with libraries like Keras and TensorFlow. ⚙️
• Bug Fixes & Stability: This version includes essential fixes and improvements that enhance overall stability, ensuring smoother execution of your ML workflows. 🐛✅
• Optimized Performance: With core runtime improvements, Python 3.11 delivers faster execution and better resource management during data processing and model training. ⚡💻

Adopting Python 3.11 is crucial for building robust, efficient applications in data science and deep learning.

2. Create the Python virtual environment

   # Run the following command to create a virtual environment in the project directory:
   py -3.11 -m venv venv  

3. Activate the virtual environment

   # Windows (using Command Prompt):
   venv\Scripts\activate
   
   # Windows (using PowerShell):
   .\venv\Scripts\activate.ps1
   
   # macOS and Linux:
   source venv/bin/activate

4. Install the dependencies

   # Run the following command:
   pip install -r requirements.txt

5. Update dependencies

   # Run the following command to update the requirements file:
   pip freeze > requirements.txt

   NT

   # Run this if the requirements file appears with strage values
   pip freeze | Out-File requirements.txt -Encoding utf8

6. Models ⚙️

🖼️ Optical Character Recognition (OCR) 📸🔠

STR model [1] trained and evaluated on the IIIT-5K Words dataset [2].
Detects and transcribes words in images with varied fonts, sizes, and noise levels, producing a clean, ordered text string.

💬 Sentiment Analysis ❤️🖤

Bidirectional LSTM RNN [3][4] trained on the Twitter Sentiment Dataset [5].
Classifies each extracted fragment as positive 😊, negative 😞, or neutral 😐, revealing the underlying intent and tone.

---

7. Application 🚀

The final app ties both models into a simple pipeline: upload an image, extract its text, then analyze its sentiment.

Usage

1. Example of usage

# In the root of the project
python -m app.main 

Then wait a litle bit to show the main menu

? Bienvenido, ¿qué desea hacer? (Use arrow keys)
 » 1. Cargar imagen
   2. Probar con un ejemplo
   3. Instrucciones
   4. ¿Quiénes somos?
   5. Informacion de los modelos
   6. Salir

Video of example of correct usage

8. Architecture

Contributions

Hualong Chiang 📖

Alfredo Fung 📖

Daniel Bortot 📖

Juan Perdomo 📖

Gabriela Martinez 📖

License

This project is under Apache license. See the LICENSE file for more details.