Edge-based text localization from natural image by classifying on the basis on the pixel distribution
Tetsuya Miyoshi, Koji Takasu · 한국지능시스템학회 국제학술대회 발표논문집 · 2009
The text information extraction(TIE) is an important task in the computer vision and nowadays challenging task for general purpose to perform in various natural condition. This paper describes the text localization method from natural images and the performance of the proposed method. The proposed method consists of three procedures, an edge extraction method, a detection of the text location based on the closed edges, and a classification of extracted candidates as text. The distributions of the white and black pixels in text areas are mixed then the proposed information measure is relatively large, whereas the measure is relatively small in the non-text. In the classification procedure, extracted candidates are classified by the information measure. The performance of proposed method is evaluated with the two examples.