Video Text Processing Method Based on Image Stitching
Tianxiang Zhou, Ke Wang, Jun Wu, Ruifeng Li · 2019
This paper studies the video of text with large field of view, solves the problem of text acquisition when a single frame cannot obtain complete text information. This paper proposes a method for acquiring video text panoramas, which automatically processes video text content by natural scene text positioning, text tracking, image stitching and other methods. First, the text is detected using the YOLO target detection framework. Second, use ECO tracking method to keep track of text. Finally, the local matching and global stitching method are used to obtain a text panorama. Experiments show that the proposed method can automatically process video and obtain high-resolution panoramic images with text parts, which greatly improves the text processing speed and effect of video content.