Research on Image Translation and Image Quality Evaluation Based on Generative Adversarial Networks

JIA CHANG, YU-DE WANG, YAN-NI JI

Abstract


How to correctly evaluate the quality of image translation is an important research topic. In this paper, the effects of the hyperparameters and algorithm optimization methods of Pix2Pix model on image translation quality are studied experimentally, and the model parameters and algorithm optimization methods are determined. The average subjective score, peak signal-to-noise ratio and structural similarity of image translation effects are proposed. And other subjective and objective indicators. On the CUFS face database, the image is translated based on the Pix2Pix model and the image translation results are evaluated. Analyze the results of image translation evaluation. The optimization method of image translation algorithm chooses Adam. When the learning rate is 0.001, the subjective index of image translation averages above 4 points, PSNR reaches 14.35, SSIM reaches 0.58, L1 loss reaches 29975251 and above, and cosin is above 0.97, are better than other methods. Finally, the validity of the image quality assessment indicators in this paper is verified on the conditional generation confrontation network model

Keywords


Image Quality Assessment, Image Translation, Pix2Pix Model, Evaluation Index.Text


DOI
10.12783/dtcse/cmso2019/33599

Full Text:

PDF

Refbacks

  • There are currently no refbacks.