Scene Text Recognition without Text Region Estimation

2015 Fiscal Year Final Research Report

• Project/Area Number: 26540124
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Soft computing
• Research Institution: Tohoku University
• Principal Investigator: Waizumi Yuji 東北大学, 情報科学研究科, 准教授 (90333872)
• Co-Investigator(Renkei-kenkyūsha): TANAKA KAZUYUKI 東北大学, 大学院情報科学研究科, 教授 (80217017)
• Project Period (FY): 2014-04-01 – 2016-03-31
• Keywords: 文字認識 / 輪郭線 / 領域推定 / 情景画像

Outline of Final Research Achievements

The purpose of this work is building an algorithm for scene text recognition without text region estimation. Almost previous works for scene text recognition consist of two recognition stages, that is text location detection and character recognition. Undetected texts at the text location detection stage are never recognized in character recognition stage. This is a serious drawback of the two stage recognition algorithm.
In this work, we propose a new scene text recognition algorithm without text location detection stage to overcome the drawback. The proposed algorithm considers an scene image as a set of closed curves and all closed curves are a part of a letter. By applying score function to all closed curves in a scene image, the proposed method can reduce un-recognition for the undetected regions.

Free Research Field

パターン認識