Video Text Location And Recognition Based On Deep Learning Algorithm

With the continuous innovation of computer technology, and the rapid development of multimedia and Internet, the approaches of video producing are varied, and video resources become richer and richer. How to extract the information we need from those video become more and more important, which has become a hotspot of multimedia research. Text is an auxiliary in the video, It plays an important role in the video understanding and contains a wealth of semantic information. Therefore, It is very meaningful for us to locate and recognize the text in video. However, the background in video is complex in general which causes OCR technology can not directly identify the text in video for a better recognition result recently.This paper is mainly about how to locate the text region in video and identify it.Firstly, This paper had a research on the characteristics of the Gabor filter, and Studied the response characteristics for text texture. It has analyzed the properties of a sinusoidal plane wave and characteristics of Gaussian function as well, which provided a method for extracting text features by using Gabor filter, and given the response of the texture features of the text in four directions. Secondly, This paper proposed a method of using deep learning in the process of text location in video to create a Deep Belief Network. To deal with the image texture features in four directions by using Gabor filter to realize the positioning of the text through the construction of the network. And it used a method of the morphology processing to corrode and expand, open and close operation for the text area in video that is located.In this way, It can remove the noise and outliers, and fill the empty area,which made the location text area more accurate.Finally, It will extract the image binarization, character segmentation and normalization and feature extraction after the morphological processing of the text area. Therefore, The text area can beeffectively identified in OCR and the recognition rate of the video text is improved as well.The text shows that, Text location method in this paper can accurately locate the text region in video. It can be effectively identified through OCR. So, The method has a certain theoretical significance and practical value.