🧠 Amazon ML Challenge – Product Info Extractor

A computer vision-based solution developed for the Amazon ML Challenge that extracts structured product information (like name, size, quantity, weight, dimensions) from product images using OCR and image processing.

This project demonstrates how to extract high-confidence text from noisy or complex packaging images, similar to those used in e-commerce, and convert that into useful metadata.

---

📌 Challenge Context

In the Amazon ML Challenge, one of the real-world problems was to extract key product details directly from images of product packaging.
These details often include:

• Product Name
• Quantity (e.g., "1kg", "500ml")
• Nutritional Info or Dimensions
• Brand and Category Keywords

---

🧠 My Solution

Using a combination of EasyOCR, PyTesseract, and OpenCV, I created an OCR pipeline that:

• Preprocesses and resizes images for clean recognition
• Extracts text using EasyOCR
• Highlights the text on the image using bounding boxes
• Applies regex and fuzzy logic (e.g., difflib) to classify product attributes
• Handles multiple formats (vertical/horizontal packaging)

---

🛠 Technologies & Libraries Used

Tool	Purpose
EasyOCR	OCR engine for high-accuracy text
PyTesseract	Alternative OCR engine
OpenCV	Image pre-processing & annotation
PIL	Image loading & format conversion
Regex	Pattern matching in extracted text
difflib	Fuzzy string matching
Requests	Handling image URLs (Amazon S3 etc.)
Matplotlib	Visualizing OCR output

---

📸 Sample Outputs

You can find more samples in the /sample_images/ folder.

---

🧪 Code Snapshot

reader = easyocr.Reader(['en'], gpu=False)
results = reader.readtext(np.array(img))
for (bbox, text, prob) in results:
    print(f"Detected: {text} with {prob:.2f} confidence")

pip install -r requirements.txt
Or run through Jupyter Notebook (notebooks/Text_Extraction.ipynb)

📌 Key Features ✅ Extracts clean text from product images ✅ Highlights detected text with bounding boxes ✅ Classifies extracted terms using regex & string matching ✅ Handles remote URLs or local images ✅ Well-commented, production-level Python code

📌 Future Enhancements 🔍 Add Named Entity Recognition (NER) to tag info like brand, size, category 📊 Store results into PostgreSQL or CSV 🧠 Add feedback loop for improving accuracy using annotation

👨‍💻 Author Danish Shaikh 🎓 BCA Student | GDG Operations Head | Data & AI Enthusiast 📧 danish89761@gmail.com 🔗 LinkedIn

ocr easyocr pytesseract computer-vision amazon-ml-challenge product-image image-processing python regex fuzzy-matching machine-learning