IDGAN: Information-Driven Generative Adversarial Network of Coverless Image Steganography

Traditional image steganography techniques complete the process by embedding secret information into cover images, but steganalysis tools can easily detect detectable pixel changes that lead to leakage of confidential information. The use a generative adversarial network (GAN) makes it possible embed using combination and noise in generating images achieve steganography. However, this approach is usually accompanied issues such as poor quality low capacity. To address these challenges, we propose model based on novel information-driven (IDGAN), which fuses GAN, attention mechanisms, interpolation techniques. We introduced an mechanism top original GAN improve accuracy. In generation model, replaced some transposed convolution operations with for better dense images. contrast traditional steganographic methods, IDGAN generates containing without utilizes GANs embedding, thus having anti-detection capability. Moreover, uses details clarity optimizes effect through algorithm. Experimental results demonstrate achieves accuracy 99.4%, 95.4%, 93.2%, 100% MNIST, Intel Image Classification, Flowers, Face datasets, respectively, rate 0.17 bpp. effectively protects while maintaining high quality.