Isru21/Amharic-Braille-Recognition-model

This model employs Python programming and leverages a Convolutional Neural Network (CNN) architecture to accurately identify and classify images of Amharic alphabet characters. --- This document provides an overview of each code file in our project, facilitating easier navigation and understanding. ### AmharicBrailleRecognitionModelTrain: This is the core project file where we - Train the Amharic Braille recognition model. - Save the trained model. - Provide new data to the model for prediction. - Essentially, all key operations are handled in this file. ### BrailleAugmentation: This code file focuses on augmenting the dataset. Here, we - Enhance the original dataset images by adjusting brightness and darkness for more diverse training. - Shift the images by 9 degrees multiple times. - Rotate the images by 9 degrees multiple times. ### ImageCrop: This part of the code processes the training dataset to - Crop each image to dimensions of (96x170). - Convert images to grayscale, turning them into black and white based on a threshold value of 128. ### AmharicBrailleLoadModels: This file contains our most accurate model for easy demonstration without retraining. Note: - The showcase works only on our local PC as the model is saved locally. - You can achieve the same results by saving the model you train in the `AmharicBrailleRecognitionModelTrain` file. ### AmharicBrailleChartPrepare: This code is used to print the same pattern multiple times, preparing the physical chart that we manually printed and colored. ---