End-to-End Text Recognition with Hybrid HMM Maxout Models

By Ouais Alsharif, Joelle Pineau. 2013

118 citations recorded by the source corpus.

The problem of detecting and recognizing text in natural scenes has proved to be more challenging than its counterpart in documents, with most of the previous work focusing on a single part of the problem. In this work, we propose new solutions to the character and word recognition problems and then show how to combine these solutions in an end-to-end text-recognition system. We do so by leveraging the recently introduced Maxout networks along with hybrid HMM models that have proven useful for voice recognition. Using these elements, we build a tunable and highly accurate recognition system that beats state-of-the-art results on all the sub-problems for both the ICDAR 2003 and SVT benchmark datasets.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Natural Language Processing

Citations in the corpus

Deep Structured Output Learning for Unconstrained Text Recognition (2014, methodology): “More recent works such as Bissacco et al. (2013); Alsharif & Pineau (2014) make use of over-segmentation methods, guided by a supervised classiﬁer, to generate candidate character proposals in a single-word image, which are subsequently classiﬁed as true or false positives.”
Speeding up Convolutional Neural Networks with Low Rank Expansions (2014, background): “test set is the collection of 5379 cropped characters from the ICDAR 2003 training set after removing all non-alphanumeric characters as in [3, 41].”
Synthetic Data and Artificial Neural Networks for Natural Scene Text Recognition (2014, methodology): “We follow the standard evaluation protocol by [2, 26, 27] and perform recognition on only the words containing only alphanumeric characters and at least three characters.”
Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks (2013, methodology): “Alsharif and Pineau (2013) use convolutional maxout networks (Goodfellow et al. , 2013) to provide many of the conditional probability distributions used in a larger model using HMMs to transcribe text from images.”
Reading Text in the Wild with Convolutional Neural Networks (2014, background): “2014), or on combining both in end-to-end systems (Alsharif and Pineau 2014; Gordo 2014; Jaderberg et al. 2014; Mishra et al. 2013; Neumann and Matas 2010, 2011, 2012, 2013; Posner et al. 2010; Quack 2009; Wang et al. 2011, 2012; Weinman et al. 2014).”
Recursive Recurrent Nets with Attention Modeling for OCR in the Wild (2016, unknown): “7 - Alsharif and Pineau [2] 74.”
Reading Scene Text in Deep Convolutional Sequences (2015, background): “It has been shown that the Maxout CNN is powerful for character classiﬁcation [18, 2].”
Focusing Attention: Towards Accurate Text Recognition in Natural Images (2017, unknown): “6 − Alsharif and Pineau [2] − − − 74.”
TextBoxes++: A Single-Shot Oriented Scene Text Detector (2018, unknown): “taset. The lexicon settings are the same as for ICDAR 2015 Incidental Text dataset. Methods SVT spotting SVT-50 spotting IC13 spotting IC13 end-to-end strong weak generic strong weak generic Alsharif [64] – 0.48 – – – – – – Jaderberg [8] 0.56 0.68 – – 0.76 – – – FCRNall+ﬁlts [27] 0.53 0.76 – – 0.85 ”
Deep Convolutional Neural Networks for Efficient Pose Estimation in Gesture Videos (2014, background): “Many recent works have demonstrated the power of ConvNets in a wide variety of vision tasks – object classification and detection [7, 13, 17, 23], face recognition [20], text recognition [1, 8, 9], video action recognition [12, 18] and many more [6,14,16].”
An End-to-End Trainable Neural Network for Image-Based Sequence Recognition and Its Application to Scene Text Recognition (2015, unknown): “Alsharif and Pineau [38] ‘ @ ‘ @ -”
Deep Features for Text Spotting (2014, methodology): “Empirically, maxout and dropout were found to be essential Wang [45] 76.0 62.0 57.0 Mishra [31] 81.8 67.8 73.2 Novikova [36] 82.8 - 72.9 Wang & Wu [47] 90.0 84.0 70.0 Alsharif [6] 93.1 to achieve this performance.”