Design And Implementation Of Text Detection And Recognition System For Natural Scenes

With the popularity of smart phone and other Internet devices and the development of 4G network and other wireless high-speed approach,it makes the way that people attain information more and more various,and convenient.The proportion of image in information media is becoming larger and larger and has extremely exceed the traditional form of text.The text within an image contain much valuable information.Traditional text detection and recognition methods mainly focus on how to handle scanned documents,bills and unified format text and merely involve text in natural scenes.Accurate detection and recognition of text in nature scenes has important significance on semantic comprehension of images.The work of this paper is based on a research project from the company the author internship.The research mainly focuses on detection and recognition of text contained in images in natural scenes.The author independently completed the following work:Design and implementation of text detection and recognition system for natural scenes.The author uses the knowledge about computer vision,and uses the algorithm of pattern recognition and deep learning to detect and recognize text in natural scenes.And the effectiveness of the approach is verified by experiments.The system include image pre-processing,text detection,text recognition and other functional modules.In the image pre-processing module,the input image is gray processed and processed to enhance contrast.In the text feature extraction module,the proposal regions are extracted by Maximally Stable Extremal Regions(MSER)method,and the deep feature is obtained from deep neural network.In the text line spotting module,the statistical methods and the information of text orientation are used to get the accurate location of text line.In the text recognition module,a neural network for text recognition is used to recognize text,and a language model is used to rectify the result.The text detection and recognition system for natural scenes is applied to detect and recognize text information contained in images in Internet.The result of testing shows the fusion of image filtering and text detection and recognition gets better performance.Thus achieves the design goal.