We propose a computational attention approach to localize and classify characters in a sequence in a given image. Our approach combines spatial soft-attention with attention regularization and learns “where-to-look” to ...