Handwritten Digit Recognition
This is one of the most classic problems in Supervised Machine Learning. From this page you can download a simple implementation, in Matlab, of a method for this problem. The training and test sets are obtained from the MNIST Database. As descriptors we used the Histogram of Oriented Gradients (HoG), without normalization. The KNN algorithm was used for classification. We obtained an accuracy of 0.96 for K = 5, 1000 training samples per class and 100 test samples per class, where the HoG was computed in cells of size 4x4, with histograms of 8 bins.
Files that you could modify in the
provided implementation are
computehog.m,
hog.m and classify.m.
Use at your own risk.