faces-and-digits-detector

In this project we implemented three different classification algorithms: Perceptron, Naive Bayes, and k-Nearest Neighbors. Each algorithm was trained using a provided training data set.\footnote{Due to time constraints, only 10 percent of the digits test dataset was used for testing each algorithm. 100 percent of the faces dataset was used to test each algorithm.} The training data set was composed of two categories, faces and digits. For each training image we defined a set of features based on the number of black pixels that occurred in certain regions of the image. Using these features, we implemented each classifier for each algorithm: one for classifying faces and the other for the digits 0-9. Finally, we compared the performance of these algorithms using the mean accuracy and standard deviation of each algorithm calculated using five separate runs. Download files and run demo.py or main.py for a demonstration.