Degraded Document Images

 in order to provide intelligent calligraphy evaluation assistance system to cope with the processing conditions of calligraphy word documents under poor lighting conditions, we jointly established our own data set with a calligraphy teaching company, which are all written on the grid paper, and stored in electronic devices by scanning or photographing, etc., and then split to single-word pictures by using the segmentation method [26-27]. In order to simulate the user's shooting conditions under different lighting environments, we randomly selected 20 single-character pictures, and took multiple sets of photos as a test set by fixing the camera position and applying different light sources on the same character, and since the photos taken are only the light source position changed, it is considered that the photos of the same calligraphic character with different lighting conditions share the same GT image, and the standard GT image can be obtained from the picture of the calligraphy character under uniform illumination by binarization algorithm and manual fine-tuning. Finally, we create a test set of 100 images, which contains 20 images under uniform illumination and 80 images under random light conditions. Some of the dataset images are shown in Fig. 9.