EMNIST: an extension of MNIST to handwritten letters

摘要: The MNIST dataset has become a standard benchmark for learning, classification and computer vision systems. Contributing to its widespread adoption are the understandable intuitive nature of task, relatively small size storage requirements accessibility ease-of-use database itself. was derived from larger known as NIST Special Database 19 which contains digits, uppercase lowercase handwritten letters. This paper introduces variant full dataset, we have called Extended (EMNIST), follows same conversion paradigm used create dataset. result is set datasets that constitute more challenging tasks involving letters shares image structure parameters original allowing direct compatibility with all existing classifiers Benchmark results presented along validation process through comparison on converted digits digits.