TernaK / Alphabets

This is a single layer neural network which is used to classify images of alphabets.

Alphabets

This is a two-layer neural network (768 input, 64 hidden, 26 output) used to classify images of alphabets.

Training

Tensorflow.

Dataset

4096 grayscale images of scrabble tiles for each letter of the alphabet. Small variations were created by moving the tiles around and rotating them slightly.

Files

• dataset.cpp uses OpenCV to capture the data.
• readdataset.py uses OpenCV to read the images and preprocess them by adaptive thresholding.
• classifier.py uses Tensorflow to train the network.
• networktester.py models the network using the weights and biases learned and can be used to test input images.
• generatevars.py exports a C/C++ header file with the network weights and biases.