Handwritten digits recognition using OpenCV

The automated recognition of handwritten digits is a largely studied problem which connects the fields of Computer Vision and Machine Learning and has many applications in real life. In this project, I detail an introductory investigation of the performance of classification in several contexts. Namely, relying on the OpenCV implementations of k-Nearest Neighbor, Random Forests, and Support Vector Machines classifiers, I compare the error rates obtained using several preprocessings of the dataset as well as various choices of features. The best obtained error rate on the MNIST dataset [2] is 0.81%, obtained with a Support Vector Machines classifier with a Gaussian kernel, using deskewing and blurring as preprocessing and using as features a combination of pixel intensities and histograms of orientations of gradients.