2014/06/09 by Max Jaderberg, Jaderberg, Max, Karen Simonyan +5 · 56 citations
Computer Science · #Handwritten Text Recognition Techniques #Natural Language Processing Techniques #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1406.2227
In this work we present a framework for the recognition of natural scene\ntext. Our framework does not require any human-labelled data, and performs word\nrecognition on the whole image holistically, departing from the character based\nrecognition systems of the past. The deep neural network models at the centre\nof this framework are trained solely on data produced by a synthetic text\ngeneration engine -- synthetic data that is highly realistic and sufficient to\nreplace real data, giving us infinite amounts of training data. This excess of\ndata exposes new possibilities for word recognition models, and here we\nconsider three models, each one "reading" words in a different way: via 90k-way\ndictionary encoding, character sequence encoding, and bag-of-N-grams encoding.\nIn the scenarios of language based and completely unconstrained text\nrecognition we greatly improve upon state-of-the-art performance on standard\ndatasets, using our fast, simple machinery and requiring zero data-acquisition\ncosts.\n