Optical Character Recognition
A projects that use neural network to train a model for optical character recognition (OCR).
Description
This project trains a multi-layer neural network that converts hand-written text to electronic format, which is known as optical character recognition (OCR). In this project, I assume that training characters have already been cut and shrunk to a uniform size(28x28 pixels), and each of them is associated with a label that indicats which character it is.
Features
Different network layouts and activation functions could be applied to compare how different neural networks perform.
Produce false-negative and false-positive images of result that could be used to visually evaluate and debug model.
Produce image of trained model by showing weight on each pixel that could be used to visually evaluate and debug model.
(Example for number characters 1-9 is shown)