A text reading algorithm for natural images
Image and Vision Computing
| Hi-index | 0.00 |
As auto-focus camera is equipped in some mobile devices, images taken from these cameras are with higher resolution and clarity currently, this provides the possibilities for character recognition in mobile devices. This paper proposes a novel and robust character recognition methods for natural scene images such as images of book cover, road signs, billboard and packing boxes. First, the text areas in an image are located by dividing to sub windows. Characters are then extracted by the connected-component (Co-Co) extraction and selection method. Finally, Fisher Component Analysis (FCA) is adopted in classifying the extracted characters. The proposed methods with a prototype system are implemented and deployed to a mobile with auto-focus camera proving the recognition result with short processing time. Testing images include pictures taken from our mobile device and ICDAR 2003 Robust Reading Competition. Where the text locating evaluation method proposed in the ICDAR 2003 competition is used to evaluate our method, and a f score=0.58 is gained in our experiment.