EdUarD0110/armenian_letters_classifier

🔤 Armenian Letter Classifier (Both Uppercase & Lowercase)

This project provides a trained machine learning model that can classify Armenian letters, distinguishing between uppercase and lowercase characters from the Armenian alphabet.

The model is designed to recognize characters from image inputs and can be used in OCR pipelines, educational tools, or Armenian language processing systems.

🧠 Model Overview

• Parameter count: 1.3M
• Type: Convolutional Neural Network (CNN)
• Input: Grayscale image of a single Armenian character (64x64)
• Output: One of the 78 classes:
  • 39 uppercase Armenian letters (Ա–Ֆ)
  • 39 lowercase Armenian letters (ա–ֆ)
  • Model's accuracy score is 0.95 (95%)

📦 Files

• mashtocimg.h5 – Trained Keras model (HDF5 format)
• label_map.json – Mapping of class indices to Armenian letters

🧾 Dataset

The model was trained on a custom dataset of handwritten and printed Armenian characters.

• Total samples: 70,000+
• Classes: 78 (uppercase and lowercase)
• Balance: Roughly equal samples per class
• Format: 64x64 images

🚀 Usage

Load the model

from tensorflow.keras.models import load_model
import numpy as np
from PIL import Image

model = load_model("mashtocimg.h5")