Classification of handwritten digits from the MNIST data set.

This project uses the unsupervised learning algorithm Kmeans to classify the images, therefore, for learning the classifier it uses only the training data set.

The program applies the Kmeans clustering algorithm using k clusters, first using random initialization (random values in [0,1]) and then using a single image that found from each label.

The program determines which of the clusters corresponds to which digit and assigns a digit to a cluster using the most common label in it.

Estimation of the label using closest centroid.