使用计算机视觉和神经网络的国际象棋游戏字母和数字分类器matlab代码.zip

# letersnumbers-classifier Letters and numbers classifier in real time for chess games using computer vision and neural networks. ## Overview This project tries to create a way to play chess by showing letters and numbers cards (describe in de folder letters_numbers) in real time, which will return the registered move (for example 'b1d1'), so a chess engine could play and move that piece. ## Methodology We first trained a neural network model (Feed Forward) using 100 images of each letter and number that were obtained recording a webcam video and storing each frame (in the video we move each card a bit to ensure the network will generalized better each one). The model should do a good work with any similar letter or number card. ### Preprocessing Each image was binarized, eroded and dilated (classical computer vision techniques). After that, it was cropped to get a new image getting rid of the background, so we just have the letter or number, and resized the image to 30x30 pixels. Here is an example:  ### Feature Extraction Now we will extract the main features of each letter and number in order to use it as an input. We will be using the 7-segment display concept, that basically divide the image in 7 parts as shown below for every possible card:  And we take the amount of letter/number in that section, that will be use as input for our neural network.  ### Training We will use those inputs to train a Feed Forward network with an input layer with 7 neurons, a single hidden layer with 10 neurons and a output layer with 8 neurons (since we will be using one network to classify letters and another for numbers). The split will be: 80% (80 images of each letter/number) for training and 20% (20 images of each letter/number) for testing so we can know how good our network is, but after that we will use every image to train the model to get as much examples as possible. ### Results and conclusion The predictions were fast enough to use then in real time, and good enough to classify each card in good conditions, but also with a bit of rotation added.   # MATLAB dependencies - Deep Learning Toolbox - Image Processing Toolbox - Statistics and Machine Learning Toolbox - Computer Vision Toolbox