作　　者： ;

机构地区： 广东工业大学

出　　处： 《计算机科学与应用》 2020年第3期456-463,共8页

摘　　要： 针对实际中的图标设计流程过于漫长和复杂的问题,参考实际图标设计流程提出了基于生成对抗网络模型的图标生成模型。首先,将用户所给出的需求转化为条件特征与噪声数据一起传入生成器中生成图标,接着将生成的图标与真实的图标一起输入鉴别器中,鉴别器鉴别图标的真假以及图标是否满足需求特征,最后二者经过迭代训练达到平衡,能够生成以假乱真的且符合需求的图标。实验表明,这种模型能够快速地生成多样的图标,且生成的图标能够满足指定的需求。 Aiming at the problem that the icon design process is very long and complicated in the real world, with the reference of actual icon design process, an icon generation model based on the generative adversarial network model is proposed based on the actual icon design process. First, the requirements given by the user are converted into conditional features and passed into the generator to generate icons together with the noise data, and then the generated icons and the real icons are the input of the discriminator. The discriminator evaluates the icon is real or not and the icon meets the required features. Finally, the generator and discriminator come to a balanced state through the iterative training. The experiment shows that the model can quickly generate a variety of icons and the icons can meet the requirements given by user.

关 键 词： 生成对抗网络 深度学习 图标 图像生成

领　　域： []