A Two-Stage Emotion Generation Model Combining CGAN and pix2pix

Abstract

Computer vision has made significant advancements in emotional design. Designers can now utilize computer vision to create emotionally captivating designs that deeply resonate with people. This article aims at enhancing emotional design selection by separating appearance and color. A two-stage emotional design method is proposed, which yields significantly better results compared to classical single-stage methods.. In the Radboud face dataset (RaFD), facial expressions primarily rely on appearance, while color plays a relatively smaller role. Therefore, the two-stage model presented in this article can focus on shape design. By utilizing the SSIM image quality evaluation index, our model demonstrates a 31.63% improvement in generation performance compared to the CGAN model. Additionally, the PSNR image quality evaluation index shows a 10.78% enhancement in generation performance. The proposed model achieves superior design results and introduces various design elements.This article exhibits certain improvements in design effectiveness and scalability compared to conventional models.