基于CycleGAN的人脸素描图像生成

Face Sketch Image Generation Based on CycleGAN

CycleGAN是一种基于生成对抗网络的衍生模型,可以在缺少成对训练图像的条件下实现两个具有不同风格的图像域之间的相互转换。由于收集大量成对的人脸图像和素描图像存在较大的难度,并且针对人脸素描图像生成任务中存在的图像细节模糊和低清晰度的问题,提出一种改进的CycleGAN模型。通过引入基于注意力机制的残差模块,让CycleGAN的生成器模型可以更加有效地学习不同通道特征和人脸图像中不同区域的重要程度,降低人脸图像中无用信息对生成模型的影响,从而提升生成的人脸素描图像的质量。通过对比实验发现,使用基于注意力机制的CycleGAN模型生成的素描人脸图像质量较好,且更完整清晰地保留了较丰富的面部特征信息,优于CycleGAN和DualGAN模型,充分证明了基于注意力机制的改进CycleGAN模型的有效性。

CycleGAN is a derivative model based on generative adversarial network,which can realize the mutual conversion between two image domains with different styles in the absence of paired training images.Because it is difficult to collect a large number of pairs of face images and sketch images,in order to solve the problem of blurred image details and low definition in the task of generating face sketch images,an improved CycleGAN model is proposed.By introducing the residual module based on the attention mechanism,the CycleGAN generator model can learn the importance of different channel features and different regions in the face image more effectively,reducing the impact of useless information in the face image on the generation model,thereby improving the quality of the generated face sketch image.Through comparative experiments,it is found that the sketched face image generated by the CycleGAN model based on the attention mechanism is of better quality,and retains more complete and richer facial feature information,which is better than the CycleGAN and DualGAN models.It is fully proved that the CycleGAN model based on attention mechanism is effective.