SSGAN: generative adversarial networks for the stroke segmentation of calligraphic characters

At present, the Chinese government is encouraging people to learn calligraphy; however, an automatic evaluation method for the results is not available. Calligraphy evaluation is challenging, because calligraphic characters are complex graphics composed of strokes as basic units. Therefore, it is necessary to consider not only the structure of the whole character but also the details of each stroke. To address this problem, we propose an automatic stroke segmentation method for calligraphic images as the foundation for the subsequent evaluation task. Specifically, we treat the stroke segmentation problem as an image to image translation problem and propose the stroke segmentation generative adversarial network (SSGAN) algorithm. Unlike the existing approaches that do not exhibit a satisfactory performance or are not suitable for use in practical projects, the proposed approach can efficiently obtain accurate results. The SSGAN enables the simultaneous generation of all the strokes of a calligraphic image, based on a multistroke tensor training strategy. Moreover, we specifically design a ResU-Net structure with embedded attention modules to enhance the accuracy of the results. The experimental results demonstrate the superiority of the proposed method over the existing state of the art models.