基于Pix2Pix的人脸素描图像生成方法研究  

Research on the generation method of face sketch images based on Pix2Pix

作　　者：陶知众 王斌君[1] 崔雨萌 闫尚义 TAO Zhizhong;WANG Binjun;CUI Yumeng;YAN Shangyi(School of Information and Cyber Security,People′s Public Security University of China,Beijing 100038,China)

机构地区：[1]中国人民公安大学信息网络安全学院,北京100038

出　　处：《智能计算机与应用》2022年第12期1-7,15,共8页Intelligent Computer and Applications

基　　金：网络安全新业态视角下的关键技术风险分析及防控对策研究(20AZD114)。

摘　　要：鉴于Pix2Pix在图像风格转换等图像翻译任务中存在细节丢失、生成图像模糊等问题,无法满足当前人脸素描生成任务的目标要求,提出一种改进的Pix2Pix模型。通过引入基于自注意力机制的残差卷积模块,让Pix2Pix的生成器和鉴别器在训练过程中能够为人脸图像的不同区域和通道赋予不同的权重,从而提高生成的人脸素描图像的质量,并且对Pix2Pix生成器的损失函数进行改进,使其生成的人脸素描图像更具有手绘风格。同时,针对生成对抗网络训练困难的问题,对原Pix2Pix的训练方法进行了改进。通过与Pix2Pix和CycleGAN对比,使用改进的Pix2Pix模型在训练过程中损失函数收敛更快、收敛过程更稳定,且生成的人脸素描图像在细节保留、轮廓清晰度等方面优于原Pix2Pix等模型,验证了改进Pix2Pix模型在人脸素描生成任务中的有效性。Considering that Pix2Pix has problems such as loss of details and blurred generated images in image translation tasks such as image style transfer,it cannot meet the current target requirements of face sketch generation tasks.An improved Pix2Pix model is proposed.By introducing the residual convolution module based on the self-attention mechanism,the generator and discriminator of Pix2Pix can assign different weights to different regions and channels of the face image during the training process,thereby improving the quality of the generated face sketch image,and the loss function of the Pix2Pix generator is improved to make the generated face sketch images more hand-drawn.At the same time,in view of the difficulty of training the generative adversarial network and the instability of the convergence process,the training method of the original Pix2Pix is improved.Through comparative experiments,it is found that the loss function of the improved Pix2Pix model converges faster and the convergence process is more stable during the training process,meanwhile the generated face sketch images are better than the original Pix2Pix and other models in terms of detail retention and outline clarity.The research verifies the effectiveness of the Pix2Pix model on face sketch generation tasks.

关 键 词：人脸素描生成 图像风格转换 生成对抗网络 自注意力机制 Pix2Pix 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]