Robust Text Binarization Based on Line-Traversing for News Video
學年 99
學期 1
出版(發表)日期 2011-01-01
作品名稱 Robust Text Binarization Based on Line-Traversing for News Video
著者 Yen, Shwu-Huey; Chang, Hsiao-Wei; Wang, Chia-Jen; Wang, Chun-Wei
單位 淡江大學資訊工程學系
出版者 Faisalabad: A N S I Network
著錄名稱、卷期、頁數 Information Technology Journal 10(8), pp.1527-1535
摘要 This study presents a robust approach to binarize the detected rectangular text regions (text boxes) on news videos. The binarization problem can be traced back to 1970s but it is still challenging in news video today since the background is complicated and unpredictable. The proposed algorithm adopts the line traversing method integrating the edge information and the intensity statistics to accomplish the binarization task. First, Canny edge detector is applied on a text box. Next, the vertical line scanning from left to right of the text box is performed twice. The vertical line traverses downwards until it hits an edge pixel or reaches the bottom of the box. Similarly, the vertical line traverses upwards until it hits an edge pixel or reaches the top of the box. These traversed pixels are classified as background pixels. From the histogram of those non-background pixels, the peak intensity p and the standard deviation σ are evaluated. The threshold for text in news video is set to be T = (0, p+kσ) or T = (p-kσ, 255), depending on text polarity. In the case that the range of background intensity covers the entire intensity range of the image, the algorithm uses the temporal information of news video to remove most of the background. Moreover, the intensities of those background pixels, whose intensity is similar to the text pixels, are replaced by 255 or 0, depending on the text polarity. Finally, a binarization is performed in this modified text box. Notice that the proposed method is parameter-free, has no limitation on the text polarity and can handle the case of similar intensity in background and text for news video. The method has been extensively experimented on text boxes from various news videos, historical archive documents and other different documents. The proposed algorithm outperforms the well-known methods such as Otsu, Niblack, Sauvola, etc., in speed, precision and quality.
關鍵字 text extraction;binarization;canny edge;text polarity;otsu;niblack
語言 en_US
ISSN 1812-5638;1812-5646
期刊性質 國外
收錄於 EI
通訊作者 Chang, Hsiao-Wei
國別 PAK
出版型式 ,電子版,紙本

機構典藏連結 ( )