Performance Analysis Anime Character Generation Based on DCGAN Model

Today's widespread animation culture and growing interest of the public have contributed to the development of a variety of animation works and characters. However, in many cases, the artists have to invest a lot of time into the idea of the task to be able to deliver their work; and dealing with such a task, obviously, is a huge consumption of time and energy. Accordingly, in many cases, these artists may not have enough inspiration to create a unique character. To tackle this issue, this paper proposed a generative network, which can be utilized to randomly generate unique anime characters. The application of the whole model is divided into three steps. First, a program is adopted for face recognition and interception of the data collected on the network. From the program and the data collected, over 50,000 photos were collected that were different and contained only the faces of the characters; second, the random generation model of this DCGAN was trained for more than a week, and paused after achieving satisfactory results; third, this model can be used to randomly generate characters and output them as images. Finally, an elaborate model can be used to generate images with regular features, distinct characteristics, and similar style as the real works. In addition, a GUI was designed to help users select their favorite randomly generated photos and save them uniformly in a specified path. The created images can be used as design inspiration for the artists who lack the time to create their new inspirations.