Research On Text Detection Technology Of Natural Scene Based On Link Line

Text,as a regular symbol,carries information and plays a very important role in people's life communication.Nowadays,with the development of Internet multimedia technology,the society has entered an era of information explosion,and the Internet is full of various forms of data,such as pictures,audio,video and so on.Obtaining the information contained in these data is an important prerequisite for us to understand and explain this society.Understanding words in pictures and videos is very important for us to understand these data and information.Therefore,detecting words appearing in pictures and videos plays an important role.Text detection for natural scene is an important research field of text detection in picture video.It is different from the general document text detection technology.There are a large number of people's daily life entertainment or picture video on the network.The background of this kind of picture video is usually complex and diverse,and the shape of the text is changed.Diversiform.In addition,there are interference factors that are not clear enough and complex illumination.The whole has brought great challenges to the identification work.Unlike in the past,people designed manual features to detect text.In recent years,with the development of deep learning technology,researchers have gradually used deep learning to deal with this problem.They generally use deep convolution networks or cyclic networks to extract more robust text features.At present,the natural scene text detection algorithm based on deep learning has been able to solve most of the application scenarios well,and has some mature applications,such as license plate recognition,identification of document information and so on.However,the vast majority of text detection algorithms still aim at straight lines in natural scenes.From the earliest horizontally arranged text to the text that is now arranged in any direction,they often directly detect word or text lines,and then output the smallest matrix in the wrapped text area.There are two main research works in this paper: firstly,a text detection algorithm based on single character detection is proposed.The two is the research of algorithm for text detection in natural scenes.Different from the current mainstream text detection algorithm,this paper first detects the location of the text block unit which belongs to the text in the image,and then clustering the basic units of the text together to get the outline of the text area through the predicted link line.Finally,the text is obtained after processing.Fine text area.Because it is based on the bottom-up processing flow,the algorithm can handle arbitrary arranged images.The experiment shows that the algorithm can detect the curving characters,get better results than the mainstream algorithm on the curved array data set,and the test on the conventional horizontal data set is better than the previous algorithm.