Research On The Application Of Scene Text Recognition Based On The Fusion Of Prior Knowledge

With the rapid development of science and technology, concept of informatization has been familiar to a lot of people. More and more people pursue a kind of powerful, concise and accurate means of communication. Smart phones with cameras and other multimedia terminal products are widely widely used,which makes it easier for people to get the natural scenes in their daily life. At the same time, The text extraction and recognition of natural scene image has a broad application prospect.Combined with the prior knowledge,this thesis will take the medical laboratory test as an example.To Set up a complete set of character recognition system, we will study the template matching, text localization and recognition technology of natural scene images.The specific contents of this thesis thesis are as follows:1) Template matching algorithm based on feature is put forward and implemented in this thesis, in order to determine the corresponding template to test sample. The algorithm is used to analyze the sample and the template, and then get the layout features of the sample and the template. In the template matching, the largest similarity category is the category of the sample by computing the similarity between the sample set and each standard template. The template matching algorithm is proposed in this thesis, which can improve the accuracy of template recognition under the interference of lens distortion factors.2) In this thesis, we improve and realize the text localization algorithm,then We locate identify the text area to be identified. The improved algorithm can adaptively search for the area to be recognized, and get the best positioning coordinates of each specific item of the laboratory, and then identify the text which has been located.In this thesis, an improved text localization algorithm is proposed, which can obtain the coordinate position of the text to be recognized more accurately.3) Based on the new template recognition algorithm and the improved text localization algorithm, this thesis designs and implements a complete text recognition system based on the distributed architecture. The system includes image input, image storage, image processing,and recognition. The distributed scheme adopted in this thesis can provide the ability of parallel processing and load balancing.Test results show that the template matching algorithm and the improved text location method designed in this thesis have more than 95% of the character recognition accuracy, which has reached the expected requirements. This system can be successfully applied in engineering practice. Experimental results show that this algorithm is effective, and has certain practical value. The system can be(but not limited to) applied to the character recognition module used in the single recognition system of medical laboratory, and it is also effective for the identification of the form or the thesis with a fixed format.