Sign Text Location In Natural Scene Image And Text Location In Whiteboard Video Conference

Text extraction in images and videos is a key issue in intelligent information processing. It has received a lot of attention in the academic field of computer vision. In this paper, the current methods of text location and the practical difficulties of video conference system are discussed thoroughly. The features of sign text in natural scene images and videos of whiteboard system is explored.This paper presents a novel method for extracting sign texts in natural scene images. In this method, edge detection is accomplished by computing the weighted Euclidean distance of pixels in different color channel of the filtered color image. Then the text regions are determined by their inherent features and relation features. The experiment results indicate that the contour of text region in the binary edge image detected by our method is both complete and clear. Furthermore, our method successfully avoids the missing of strokes and adherence of characters. The veracity of detecting the text region is also improved.In consideration of the unique requirement of whiteboard video conference, an automatic note-taking method is designed and implemented. In this method, the whiteboard is located, rectification to estimate background color. Foreground cells and text cells are classified by whiteboard color. The key frame and the time stamp of strokes are obtained by counting the number of stroke cells and the time they appear on the whiteboard. This method dramatically lowers the transmitting bandwidth of video conference. Compared with existing hardware note-taking system, such as electronic whiteboard system and electronic pen, this method claims a good commercial prospect for its convenience and low-cost.