Research on Visually Rendering Technology for Images Based on Generative Adversarial Network and OpenCV

The prediction accuracy and stability of traditional image visual rendering are poor, resulting in low image rendering efficiency. This paper proposes an image visual rendering technology based on the combination of self-attention generative adversarial network and OpenCV. Firstly, the original grayscale image is input into the generative adversarial networks (GANs) for training. Then, based on the trained GANs network, a self-attention mechanism (SA) is added to build an image rendering model based on SAGAN. Finally, SAGAN model is combined with OpenCV technology to arrange the frame order and synthesize the bit rate of the sample images, so as to obtain clear color images. The results reveal that compared to GAN network, SSIM index and PSNR index of SAGAN network are increased by 20.32dB and 19.27% respectively, indicating that adding self-attention mechanism to GAN network can improve the attention of the model to important information, and the image rendering quality is significantly improved. Under the same test set, the IS value of the proposed algorithm is 57.29 scores, which is significantly higher than that of the Criminisi algorithm and the gate convolution-based GANs algorithm. In addition, the image rendering time of the proposed algorithm is only 6.54h, which is lower than that of the other two algorithms. It proves that this algorithm can ensure the classification stability of discriminator and the coloring uniformity of generator, thereby improving the accuracy and efficiency of image time rendering prediction.