Persian_Handwritten_Digit_Recognizer

Persian Handwritten Digits Recognizer Using

• Python3
• Neural Networks
• Tensorflow / Keras
• Cv2
• Numpy

Preprocessing

• Displaying the raw samples
• Normalizing the pixel values
• Applying thresholding to simplify the data
• Performing erosion and dilation to clean up broken or disconnected pixels
• Optionally applying morphological closing (can potentially reduce accuracy)

First Model

Consists of three layers: two dense layers with ReLU activation functions and one dense layer with softmax activation function.

Best accuracy: loss: 0.0695 - accuracy: 0.9792 (with closing)

Second Model

Consist of four layers: three dense layers with ReLU activation functions and one dense layer with softmax activation function.

Best accuracy: loss: 0.0652 - accuracy: 0.9847 (with closing)

Third Model

Consist of three layers: two dense layers with ReLU activation functions and one dense layer with softmax activation function.

Best accuracy: loss: 0.0541 - accuracy: 0.9880 (without closing)