A Robust and Secure Image Steganography using Convolutional Neural Networks and Transform Methods

The act of obscuring the existence of data in images is commonly referred to as steganography. Contrary to cryptography, there is no data scrambling. The original and the cover data carrying the hidden confidential data are so similar that human perceptions cannot differentiate. Steganography is an ancient and efficient way to hide and communicate confidential information. The effectiveness is measurable using various parameters. The steganography analysis method examines the captured data with various parameters in an attempt to identify the steganography. This research study presents an appropriate way to perform image steganography using a combination of techniques like the least significant bit, discrete cosine transforms, and a combined model of convolutional neural network (CNN)-Convolutional Autoencoder with Encoder-Decoder based Latent Features (CAEDLF). LSB-based steganography is an in-demand technique for concealing confidential information in the LSBs of pixel values without creating any noticeable deformity. DCT helps provide robustness by changing the structure of the secret image. The Convolution Neural Network (CNN) based convolutional autoencoder with encoder-decoder-based latent features (CAEDLF) autoencoder is used for optimizing steganography. The proposed steganography shows the best results in terms of robustness, security, and imperceptibility capabilities and also helps optimize steganography analysis.