The Text Recognition Of Street View Images

With the rapid development of computer technology, the human society shiftfrom the industrial society to information society, information technology isincreasing constantly. With the development of Internet technology, thetransmission of information between people no longer rely on paper-basedmaterials, all files could be converted to electronic documents, which requirepeople manually input the text information in their daily work. It adds a great dealof duplication of work. In order to improve the efficiency of daily work, manyscholars begin to pay attention to the text recognition technology. Meanwhile, thecomputer instead of human beings to accomplish these repetitive works on thecomputers made the equipment become more intelligent. At present, the computertext recognition technology is well established, for printed Chinese characters,Chinese characters scanned documents, the identification speed is rapid, and theaccuracy is high, which greatly improves the efficiency of our work.Mobile mapping system (MMS) is the advanced technique in modern surveyingand mapping industry, the system solves the massive spatial data storage and fastupdate issues, further extracts more useful information from these massive data. Asone of the basic data which collected by MMS, Street View images contains a largenumber of text messages, if these messages can be automatically extracted, it isvaluable to search and retrieval information and acquire the location of points ofinterests.This paper analyzes and solves text recognition problems in Street Viewimages by combining spatial information technology and text recognitiontechnology. The Street View image information in the text background is complexand the position of image is uncertainty, many texts are attached on theadvertisement plaque, and most of the texts have more artistic effect. To identify these texts, the first step is to make sure the scope of the text, this study putforward two kinds of analysis methods: mathematical morphology analysismethods and connected component analysis methods for text regional positioning.Text extraction from complex background has laid a good foundation for textrecognition.Text recognition is one kind of pattern recognition, pattern recognition theoryis an identification method proposed by researchers’ in-depth study of humanbehavior and thinking, it based on the development of computer technology. Thereare many kinds of pattern recognition. Among these, artificial neural networkpattern recognition has broken the traditional pattern recognition and it has openedup a new research direction. As the neural network based on high-speed fractionsparallel processing working principles, it has a strong self-learning ability,associative memory capacity, fault tolerance, scalability, organizational and otherfeatures. Therefore, the use of neural network text recognition method can improverecognition quality and recognize more predefined character. In this paper, singlehidden layer BP neural network text recognitio n was used for text recognition.Experimental results show that the proposed text area positioning method andneural network text recognition method can more accurately locate the text regionand the text in the text region can be identified basically.