A language-independent, open-vocabulary system based on HMMs for recognition of ultra low resolution words

Authors:
Farshideh Einsele;Rolf Ingold;Jean Hennebert
Affiliations:
University of Fribourg, Fribourg, Switzerland;University of Fribourg, Fribourg, Switzerland;Institut Informatique de gestion, Sierre, Switzerland
Venue:
Proceedings of the 2008 ACM symposium on Applied computing
Year:
2008

Citing 7
Cited 0

Fundamentals of speech recognition

Fundamentals of speech recognition
Twenty Years of Document Image Analysis in PAMI

IEEE Transactions on Pattern Analysis and Machine Intelligence
A Statistical Approach for Phrase Location and Recognition within a Text Line: An Application to Street Name Recognition

IEEE Transactions on Pattern Analysis and Machine Intelligence
Locating and Recognizing Text in WWW Images

Information Retrieval
Text Extraction from Web Images Based on A Split-and-Merge Segmentation Method Using Colour Perception

ICPR '04 Proceedings of the Pattern Recognition, 17th International Conference on (ICPR'04) Volume 2 - Volume 02
Web image indexing by using associated texts

Knowledge and Information Systems
A HMM-based approach to recognize ultra low resolution anti-aliased words

PReMI'07 Proceedings of the 2nd international conference on Pattern recognition and machine intelligence

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this paper, we introduce and evaluate a system capable of recognizing ultra low resolution words extracted from images such as those frequently embedded on web pages. The design of the system has been driven by the following constraints. First, the system has to recognize small font sizes where antialiasing and resampling procedures have been applied. Such procedures add noise on the patterns and complicate any a priori segmentation of the characters. Second, the system has to be able to recognize any words in an open vocabulary setting, potentially mixing different languages. Finally, the training procedure must be automatic, i.e. without requesting to extract, segment and label manually a large set of data. These constraints led us to an architecture based on ergodic HMMs where states are associated to the characters. We also introduce several improvements of the performance increasing the order of the emission probability estimators and including minimum and maximum duration constraints on the character models. The proposed system is evaluated on different font sizes and families, showing good robustness for sizes down to 6 points.