Secure Translation from Sketch to Image Using an Unsupervised Generative Adversarial Network with Het for AI Based Images
M. Poongodi, N. Buvaneswari, Sanjay Kumar Bose, N Maheswaran, S. Vijayalakshmi · 2023
Sketch to image translation could be a useful tool in identifying person and generating real life-like pictures out of sketches and drawings. Translation of sketches to image can come in handy in comic generation or graphic designing applications. The existing system under perform with the accuracy of the images generated and generating images with desired background image is also an important area to improve on. So, it is essential for a system to produce images with better accuracy and quality with desired background. The sketch to image translation is generally deployed using Convolution Neural networks (CNN) and Generative Adversarial Network (GAN) but the solutions often fall behind and fail to make high quality or a real life like output. While these methods focus on bringing out images resembling the given sketches, they fail to concentrate on quality of images generated. The images generated from existing Sketch to image translators and not real life-like and the generators themselves underperform in terms of accuracy. So, it is essential for a system, which can translate sketches into images without labels with better accuracy. The quality and the integrity of the sensitive images should be protected by using Homomorphic Encryption Technique (HET).