Affiliations: SpeechGear, Northfield, MN, USA | Department of Computer Science, Texas Tech University,
Lubbock, TX, USA
Abstract: Translating text in real-world images presents several challenges
such as text detection, text extraction, recognition, and translation. A
multi-lingual translation system must take fundamental differences between the
characteristics of different alphabets such as Latin, Cyrillic, Chinese,
Korean, and Arabic into account. The system presented in this paper can extract
text from real-world images with appropriate heuristics for these alphabets, as
well as de-skew, binarize, recognize, and translate them. OCR is utilized to
recognize the text, and the translation is employed using Translation Memory
and Machine Translation. Experiments were also conducted to determine the most
suitable segmentation and binarization algorithms for translation.
