Abstract:
“What we need is a machine that can learn from experience” -Alan Turing, 1947 Despite the emergence of precision and speed of single image resolution using fast and in depth neural convolution networks, one central problem remains unresolved: how do we get good texture details when we properly fix big growing objects? Recent work has focused on reducing the risk of mean squared reconstruction error. The payload ratio has high signal-to-noise ratios (PSNR) and has failed to operate at high frequencies. Therefore, we introduce SRGAN, a super-resolution (SR) Generative Adversarial (GAN) network. The most challenging task of measuring an image with high resolution (HR) from its low resolution counterpart is called super-resolution (SR). To our knowledge, this is the first draft capable of capturing a realistic picture of 4x magnification features. The mean-opinion-score (MOS) test shows hugely significant gains in perceptual quality using SRGAN. The MOS scores obtained with SRGAN are close to those of the original high-resolution images than to those obtained with any state-of-the-art method. The main purpose of the project is to build a Generative Adversarial Network using deep learning. Firstly, we train the generator network with LR images as well as adversarial network with HR images. After the training of both networks, we give LR image as input to GAN which generate the super-resolved image. The project will be useful for security surveillance and in medical diagnosis.
