KousarRaza/sindhi-hand-writing-ocr

Sindhi OCR Model

Model architecture	Model size	Language
CNN + Dense	~XX MB	Sindhi

---

📝 Description

This Sindhi OCR model is designed to recognize handwritten or printed Sindhi characters from scanned or digital images. It supports a basic set of characters and outputs the recognized text as a string.

---

🔍 Key Features

• Recognizes isolated Sindhi characters from 28x28 grayscale images.
• Pretrained and ready for inference using TensorFlow/Keras.
• Small and lightweight – easily integratable in web or mobile applications.

---

🧪 Use Case

This model is ideal for:

• Educational tools for Sindhi literacy
• OCR engines for digitizing Sindhi handwritten documents
• Research on low-resource language OCR
• Basic Sindhi text recognition pipelines

---

📥 Input

• Type: 28x28 grayscale image
• Format: NumPy array or .png, .jpg
• Shape: (28, 28, 1)

📤 Output

• Type: Text
• Format: Unicode string
• Classes: ['ا', 'ب', 'د', 'م', 'ي', 'چ']

---

🛠️ How to Use

🔗 Load the model in Python

from tensorflow.keras.models import load_model

model = load_model("sindhi_model.h5")

# Predicting from an image
import numpy as np
from PIL import Image

img = Image.open("test.png").resize((28, 28)).convert("L")
img = np.array(img).reshape(1, 28, 28, 1) / 255.0

prediction = model.predict(img)
predicted_class = class_names[np.argmax(prediction)]