生成对抗网络的素描生成方法

Sketch Generation Method Based on Generative Adversarial Network

从彩图/照片生成素描,其生成素描图像的轮廓、阴影与铅笔素描画相比不够清晰,精准度也不高.由于素描笔画类型多样,阴影结构复杂.为此,本文提出了一种人脸照片到铅笔画的转换方法,设计并实现了一个双分支训练生成对抗网络模型,其轮廓分支提高生成素描图像的轮廓清晰度;另一阴影分支处理素描的纹理与阴影部分,并且保持素描画的风格.最后,本文对不同类型的素描风格转换进行了实验,结果表明提出的方法比现有方法有更好的素描生成效果.

Sketches can be generated from color pictures or photos,but compared with pencil sketches,the outlines and shadows are not clear enough and the accuracy is not high.Due to the various types of sketch strokes,the shadow structure is complex.To this end,this paper proposes a method for converting face photos to pencil drawings,and designs and implements a new dual-branch training generative adversarial network model.The model has two branches,one contour branch improves the outline definition of the generated sketch image;the other shadow branch handles the texture and shadow parts of the sketch,keeping the sketch style.Finally,this paper conducts experiments on different types of sketch style conversion,and the results show that the proposed method has better sketch generation effect than existing methods.