Frequency aware face hallucination generative adversarial network with semantic structural constraint

In this paper, we address the issue of face hallucination. Most current hallucination methods rely on two-dimensional facial priors to generate high resolution images from low images. These are only capable assimilating global information into generated image. Still there exist some inherent problems in these methods, such as, local features, subtle structural details and depth missing final output This work proposes a generative adversarial network (GAN) based novel progressive (FH) issues present among methods. The generator proposed model comprises FH two sub-networks, assisting first sub-network leverages explicitly adding frequency components model. To encode components, an auto encoder is coefficients discrete cosine transform (DCT). add three dimensional parametric network, second proposed. uses shape 3D morphable models (3DMM) constraint network. Extensive experimentation evaluation show usefulness architecture form state-of-the-art quantitative results. • paper presents multi module GAN learns leverage image texture. Second adds information. Experiments benchmark datasets reflect superior performance over state-of-the-art.