A Realistic Image Generation of Face From Text Description Using the Fully Trained Generative Adversarial Networks

Abstract

Text to face generation is a sub domain of text to image synthesis, and it has a huge impact along with the wide range of applications on public safety domain. Currently, due to the lack of dataset, the research work focused on the face to text generation is very limited. Most of the work for text to face generation till now based on the partially trained generative adversarial network, in which the pre-trained text encoder has been used to extract the sematic features of input sentence. Then these semantic features have been utilized to train the image decoder. But in this research work, we have proposed the fully trained generative adversarial network to generate the realistic and natural images. We have trained the text encoder as well as the image decoder at the same time to generate the more accurate and efficient results. In addition to proposed methodology, we have also generate the dataset by the amalgamation of LFW, CelebA and locally prepared dataset. We have also labelled the images according to our defined classes. Through performing different kind of experiments, we have proved that our proposed fully trained GAN outperformed by generating the good quality images with accordance to the input sentence. Moreover, the visual results have also strengthened our experiments by generating the face images according to the given query.

Brain tumor detection and multi‐classification using advanced deep learning techniques

“…Nonetheless, it is hard to compare the results with the techniques reported in state of the art due to different CNN models and data set employed (Khan, Jabeen, et al, 2020). However, we still compare current techniques on brain tumor detection and classification.…”

“…For the classification of multimodal automatic brain tumors with linear contrast stretching, the transformation of learning‐based extraction functions, correntropy‐based features set, Khan, Jabeen, et al (2020) proposed a fusion of pre‐trained CNN models (VGG16–VGG19). The fused matrix was eventually fed to the extreme learning machine for brain tumor identification.…”

“…Nonetheless, it is hard to compare the results with the techniques reported in state of the art due to different CNN models and data set employed (Khan, Jabeen, et al, 2020…”

Internet of medical things embedding deep learning with data augmentation for mammogram density classification

“…The basic CNN is one input, one output and hidden layers. In hidden layers, convolution and pooling operation is completed (Ramzan, Khan, Iqbal, et al, 2020; Ramzan, Khan, Rehmat, et al, 2020; Mashood Nasir et al, 2020; Khan, Jabeen, et al, 2020; Khan, Sharif, et al, 2020; Khan, Akram, et al, 2019; Khan, Nazir, et al, 2019). The pooling operation is achieved either through max‐pooling or average pooling to reduce the computational cost and to find meaningful patterns by the subsequent layer.…”

Using genetic algorithms to uncover individual differences in how humans represent facial emotion

“…However, these techniques often rely on the secondary judgement of a skilled artist and are limited in their ability to allow users to explore and interpret resulting graphics. More recently, machine learning techniques such as generative adversarial networks (GANs) [25,26] and verbal crowd-shaping [27] have been applied to create computer-generated facial expressions and body images based on sampled parameters and descriptive labels from a population. While promising, these methods rely on unsupervised learning and classification, requiring vast training datasets and fully trained networks for optimal performance.…”