PURVEY-CE: A Complex Texture Adaptive Image Steganography Based on Channel Attention

PURVEY-CE: A Complex Texture Adaptive Image Steganography Based on Channel Attention

연구 분야: Strategies

학회: International Conference on Neural Information Processing

Most of the existing image steganography algorithms often overlook preprocessing image edges and complex textures, limiting their invisibility and spatial flexibility. Thus, this paper proposes a new image steganography algorithm—PURVEY-CE(A comPlex textURe adaptiVe imagE steganographY based on Channel attEntion). Due to the rich change of image edge and complex texture, PURVEY-CE utilizes texture complexity and color distribution to increase hiding space of images. Then, channel attention mechanism improves selection validity and embedding precision of the hidden location. Meanwhile, the embedding weights are adjusted based on channel texture complexity to enhance the accuracy of the feature map. Further, PURVEY-CE designs steganographic distortion functions from multiple dimensions to optimize steganographic image quality and model security. Experimental results show that, the PSNR of PURVEY-CE is up to 47.85 dB, it is 11.15% higher than other adaptive steganography algorithms, and the information extraction accuracy is also improved. It indicates that PURVEY-CE exhibits superior performance in embedding and adaptability in complex texture images, ensuring higher steganographic security and stability.