Study On Text Detecting From Video Images

Along with the advance of digital process, Multimedia and communications industryhave found wide application in our daily life. Digital videos and images have been as themain media, which is omnipresent nowadays and increase constantly. To meet theincreasingly diverse range of requirements of the masses of the people, the technology ofdigital video processing has made substantial progress.There are so many digital videos in everyday life, including digital cinema, networkvideo image, and medical image and supervising videos. Processing a great deal of thosemultimedia data has drawn worldwide attention and we can see great potential in this.Recently, study for text detection from video images has been a hotspot in the area ofimage processing and analysis. This thesis, focusing on text detection and segmentationfrom video images, make a remarkable research. The main content of this thesis include3aspects as below:(1) In this thesis, author firstly analysis and compares some commonly used textdetection and segmentation, and points out similarities and differences between them.(2) Researching on text detection and localization, this thesis presents a novel andeffective method to detect and locate text line from video image having complexbackground. By analyzing the complexity of background, video images has been classifiedand described as below: simple background, normal background and complex background.Different backgrounds have different adaptive method, which can assure accuracy as wellas increase efficiency. Especially, for the complex background images, we make thoroughanalysis and research, and adopt several succession video images to remove false alert.Increase the efficiency and precision of the novel method.(3) Researching on text segmentation, this thesis present a systematic textsegmentation algorithm based on commonly used text segmentation, which includesanalyze horizontal projection and vertical projection, judge the polar of the text area,multiframe verification and stroke segmentation.For the above algorithm, the thesis has gathered a large number of video images andmade a lot of testing. Testing results show that the proposed method has achieved good results in terms of efficiency and effect of the performance.