Comparison of deep conditional generative models for scanning electron microscopy image reconstruction

Le résumé fourni par la source

Image-to-image translation is a task in the field of computer vision that has been gaining importance in recent years, with the objective of transforming an image from one visual domain to another without losing coherence. This process has applications in various tasks, such as style transfer, and in different fields, including medicine. One area where image translation has great potential is materials science, where obtaining a series of images of a material using scanning electron microscopy can be a complex process. In this context, image-to-image translation emerges as an alternative for generating synthetic images of a material’s microstructure, based on specific visual information obtained by the scanning electron microscopy process. In this work, we present a comparison of the performance of different conditional generative architectures for obtaining synthetic images from edge maps. These edge maps are obtained through image pre-processing algorithms such as the Laplacian and Canny edge detectors, complemented with image manipulation and correction techniques. To evaluate the performance of the conditional generative adversarial networks, metrics such as the structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR) are employed. The results demonstrate that both the choice of edge detection algorithm and the correction techniques significantly impact the model’s ability to generate synthetic images with high fidelity.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.