Publication

Abstract : Steganography is the technique that involves hiding a secret data in an appropriate carrier. The major challenge involved in steganography is to ensure that the hidden data does not attract any attention towards it and hence works under the assumption that if the secret feature is visible, then the point of attack is evident. In this work, a novel deep learning model is designed to perform digital image steganography. The dataset used to train the model is Common Object in Context (COCO). An analysis is conducted based on batch size hyper-parameter, to evaluate the performance of the model. Also, the effect of using grayscale and color images on the evaluation metrics of the model is estimated. The analysis was orchestrated by evaluating the average Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index (SSIM) of the trained images. The analysis has produced state-of-the-art results with optimized parametric values and has boosted computational efficiency producing a promising architecture to perform steganography.