A Unified Framework for Encryption and Decryption of Images Based on Autoencoder (UFED)

Abstract

Over the last decade, many disciplines have made great strides in deep learning technologies, especially in computer vision and image processing. However, video coding based on deep training is still in its initial stage. This research work discusses the representative's work on deep learning for image/video coding, a research area since 2015. With the number of devices increasing on the Internet, we face low-cost transmission over a network and security and safety. We can't determine the accurate data size with encryption and decryption cost and amount of noise in communication. Our proposed unified framework for encryption and decryption of images based on an autoencoder (UFED) can control the cost during encryption and decryption using modern techniques like deep learning and neural network. The Autoencoder is worked as close to CNN and is trained on images and video frames to extract the image's feature. In this framework, the encoder changes the image into latent space or compressed form in a small size. We achieved the best image-compression ratio with Autoencoder over JPEG; JPEG typically achieves 10:1 compression with little perceptible loss in image quality. This research observed the accuracy of image reshaping from latent space as well. We have achieved over 97.8% accuracy on the standard quantity evaluation measure in our proposed deep learning technique, far better than previously implemented models.