A Research On Text Extraction In News Video Based On Corner

In people's daily life,it is an extremely important way to get information through watching news video in contemporary society.The accuracy of text localization and text extraction plays an important role in the retrieval based on video content.So it has become an important research direction in the field of multimedia information processing.However,the resolution of news video is too low and often embedded in the complex background,which has brought a great difficult to text localization and extraction.Next,we focus on the research of text localization and text extraction in video under complex background.A method based on corners was proposed in news video:Firstly,the existing algorithms have a problem,which is easy to generate false alarm rate under complex background in video.A multilayer filtering mechanism of text localization algorithm based on corner was proposed.In this stage,our text localization is based on corner,which can accurately describe the characteristics of the text in video and which is stable in singular contrast.Firstly,corner detection was performed.Then the multilayer filtering mechanism was carried out for text localization,which consists of corners clustering,corners horizontal projection,background filtering and heuristic rules.The isolated corner was removed by corner cluster,which can reduce the effect on the next step.Secondly,the horizontal projection can filter out the corners of none text area and get the candidate text lines area.Finally,the horizontal sliding window and heuristic rules were used to reduce the residue background text lines or pseudo text lines and get the accurate text lines images.This method can effectively remove the nose corner and accurately locate text lines of news video in a complex background.Secondly,the existing text extraction algorithm with different contrast can be not very good to be solved under complex background.A polarity judgment combine with twice binarization was proposed for text extraction.This paper's text extraction combined polarity judgment with twice binarization:Firstly,the novel polarity judgment algorithm,which effectiveness is verified through experiments and statistics,was used to determine the polarity of text lines.The polarity judgment was used as a guide to adjust the first binarization threshold when we perform the first binarization.After the first binarization,a main proportion of the image has been processed,and the rest will be processed by the twice binarization.The threshold also was adjusted by polarity judgment.Now,the binary images were produced,which has clear and integrated stroke details.Finally,the text recognition.The binary images obtained by text extraction were identified by OCR(optical character recognition)plug-in.The open source Tesseract-OCR was utilized in this paper.The feasibility and effectiveness of the proposed method were verified by experimental results in news videos,which has higher stability and accuracy even in a complex background and environment.