Video Text Locating And Extraction

Nowadays,with the rapid development of communication equipment and digital multimedia technology,bandwidth and storage are no longer the bottleneck for watching high-definition videos.Consequently,the form of Internet media has turned from text,picture into video.If it is possible to locate and extract the text in videos,rich textual information could be acquired,which is very valuable for understanding and retrieving high-level semantic information in videos.Therefore,video text locating and extraction is an active topic in the field of computer vision and artificial intelligence.In this thesis,the issues in video text locating,segmentation and extraction systems are analyzed and then an efficient video text locating and extraction system is proposed.Firstly,the input video file is decoded and video frames are sampled at fixed rate.For video frames,there are two steps in text locating.In the first step,corner response image is extracted and gray level morphological operation and adaptive threshold segmentation is used to locate the candidate textual regions.In the second step,features of stroke width histogram are computed for the candidate textual regions and machine learning methods are adopted to classify true textual regions.For the segmentation of textual regions,a method based on Fuzzy C-means clustering is used to separate the background image into layers by the color and location information of its pixels,so as to extract the text layer.The last process of text segmentation is using morphology methods to repair the skeleton of text.At last,the result of text segmentation is fed to OCR software to acquire textual information.Experimental results show that the proposed system outperforms previous algorithms in precision rate,recall rate and f-measure.