It is a database of handwritten digits contributed by different writers. It is a subset of a larger set available from NIST. The digits have been size-normalized and centered in a fixed-size image. It consists of a training set of 60.000 images and a test set of 10.000 images.