A Method for Text Detection and Rectification in Real-World Images
Satoshi Yonemoto · 2014
This paper presents a text detection method that combines with an image rectification. Since texts in natural scenes are not always observed in frontal view, image rectification is needed to robustly recognize them in OCR. A reference pixel that is part of the desirable text area is given by user since this is the easiest way to give the priors. In text detection, first, foreground pixels are extracted. Then, the text components and its structure are analyzed, constructing a graph with the text components. The image rectification is executed based on quadrangle estimation.