Paper Title
The String Detection of the Complex Format Table Image

Abstract
This study detects text string in complex format table images. It also generates large complex format table images for changes in table types, backgrounds, and text forms. This study uses YOLOv4 and U-Net for preliminary detection of a text string, and there are too many variables in complex format table images. This study analyzes the input image method to obtain better text string detection. Also, De-noise processing is performed on the detection results of YOLOv4 and U-Net, respectively. Finally, these detected results are merged as the detection of the text string. The study used 4 tables and images in complex format for experiment and performance evaluation. In the performance part of text string detection, the UNet& YOLOv4&De-noise method is adopted, and the accuracy of text string detection is the highest. It is worth mentioning that there are not many experimental images in this study. This study proposes a method to generate large training images, improving text string detection accuracy in complex format table images and improving subsequent character recognition accuracy. Keywords - Complex Format, Table Image, Deep Learning, String Detection, Image Generation