Under The Natural Scene Character Area Location And Recognition

As the development of the mobile phone and other smart mobile devices, people pay more and more attention to the acquisition and the application of text information in natural scene.Text information in natural scene is different from the printing text,which has a variety of languages, different font and size, and is affected by the complex interference of background,deformity as well as pollution. All of these bring the challenges and difficulties for the acquisition of text information in natural scene.In this paper, we focus on studying the accurate positioning and identification of the text area in the natural scene, and we mainly study on the English characters and numbers.Moreover,we also research on a special case i.e. the text area positioning and identification of railway tanker’s car number in complicated industrial environment. The text area of railway tanker’s car number is also a kind of text area in the natural scene which has the characteristic of faulted characters. Therefore,we regard the text area positioning and identification of railway tanker’s car number in complicated industrial environment as a special natural scene.Our goal is to achieve accurate identification of the railway tanker’s car number area and the separation and identification of faulted characters.In this paper, we first make comparison and summary of various text area localization methods, and then give a common method which can be applied to position the widespread English letters and digital region in natural scenes,and also can be used to position the text area of railway tanker’s car number in complex industrial environment. This method has a good positioning result for the text area, which is different from the size, has the existence of tilt and is affected by the variation of the illumination. First of all, the Maximally Stable Extremal Regions(MSER) to applied to obtain the maximum areas, and acquire the effective areas from the maximum areas. Then, we can get the triplet areas from the effective areas.The available triplet areas is connected as the candidate text areas. Finally, the Support Vector Machine(SVM) to applied to screen candidate text areas.The method has good performance in positioning of natural scene text area. In order to verify the generality of the method, it is applied to acquired the railway tanker’s car number in the complex industry environment. As the characters of the number often have the broken strokes, which must be taken in count,we apply a character splitting method which can handle the problem. Considering many kinds of the characters and the changes of the font for the natural scene, and the single characters and font for the number of the train tank,we apply different methods to recognize the characters in the two cases. For the natural scene, we apply the Tesseract-OCR to training classifier and recognize. For the number the train tank, we apply the SVM to training classifier and recognize. A large number of experiments show the satisfied results of the method.