利用DCGAN实现相似图片自动绘制的应用“图自绘”的设计与实现研究

The Design and Implementation of “Automatic Drawing” by Using Dcgan to Realize the Automatic Drawing of Similar Pictures

摘要: 绘图起源于新石器时期，古人将天马行空的幻想记录在岩石和彩陶上，直至今日绘图仍是我们生活中不可缺少的部分。大到工业制图，小到街头广告，现如今人们生活中大大小小的事情都需要接触到绘图。于是，在互联网技术的发展和基于大数据的人工智能发展下，我们对利用深度卷积生成对抗网络(Deep Convolutional Generative Adversarial Networks, DCGAN)实现相似图片自动绘制进行了应用设计与实现研究。然而深度卷积生成对抗网络需要大量的计算，对硬件的要求较高，为了降低用户的使用门槛，我们对应用进行了轻量化设计模式实现，即把计算任务交给服务器端来完成，用户端只作为需求上传与成品下载的媒介。具体来说，首先我们借助服务器端–用户端的轻量化设计理念完成了“图自绘”应用的总体框架，然后对应用中所包含的每个功能模块完成实现研究，最后根据相应的设计开发流程使用Android Studio完成应用的设计与实现。借助服务器端的快速计算与用户端的需求上传与成品下载，用户可以在弹指间获取到自己心仪的图片，为有绘图需求但无绘图技术的人群开辟了一种获取图片的新途径。

Abstract: Drawing originated in the Neolithic period. The ancients recorded their illusions on rocks and painted pottery. Today, drawing is still an indispensable part of our life. From industrial drawing to street advertising, nowadays people need to touch drawing in their daily life. Therefore, with the development of Internet technology and artificial intelligence based on big data, we study the application design and implementation of automatic drawing of similar pictures by using deep convolutional generative adversarial networks (dcgan). However, the deep convolution generation countermeasure network needs a lot of computing, and has high requirements for hardware. In order to reduce the user’s threshold, we implement the lightweight design pattern of the application, that is, the computing task is handed over to the server, and the user is only used as the media of demand upload and finished product download. Specifically, first of all, we complete the overall framework of the “self drawing” application with the help of the lightweight design concept of server-side and client-side, then complete the implementation research of each functional module contained in the application, and finally use Android studio to complete the design and implementation of the application according to the corresponding design and development process. With the help of server-side fast calculation and user-side demand upload and finished product download, users can get their favorite pictures between the snap fingers, which opens up a new way to get pictures for people who have drawing needs but have no drawing technology.