Improved GAN for Efficient Cyber Forensics

The society is now deeply connected with digital exchanges running on cyber-physical systems for their daily operations. This gives rise to cyber crime and creates a need for cyber forensics. This paper presents an innovative approach based on improved Generative Adversarial Networks(GAN) for efficient cyber forensics. The proposed system uses the existing GAN networks of SRGAN and DCGAN to generate high quality images from low quality images. We experiment with different activation functions, number of layers, nodes and network architectures to improve the accuracy achieved in super resolution of images. The system has been tested for use in cyber forensics to super resolve the images of accused in a video frame and for better detection of car number plate images taken from cameras on traffic lights and CCTV cameras. The experimental results show that our proposed system outperforms existing GAN networks in terms of PSNR value by ~2.24% and the SSIM method by ~3.59%.