TSCL:Multi-party loss Balancing scheme for deep learning Image steganography based on Curriculum learning

TL;DR

The paper tackles the challenge of balancing multiple losses in deep learning image steganography to simultaneously optimize invisibility, recoverability, and security. It introduces TSCL, a two-stage loss scheduler that uses a priori curriculum control to order learning priorities and loss dynamics control to adapt weights based on loss drops across tasks, focusing first on embedding, then decoding, and finally steganalysis resistance. Evaluations on ALASKA2, VOC2012, and ImageNet show that TSCL improves perceptual quality metrics (e.g., $L_{Encode}$ measures such as PSNR/MSSSIM/SSIM), decoding accuracy, and resistance to steganalysis compared with fixed-weight baselines. These findings underscore the practical potential of curriculum-based, dynamic loss balancing for multi-task steganography and secure data transmission.

Abstract

For deep learning-based image steganography frameworks, in order to ensure the invisibility and recoverability of the information embedding, the loss function usually contains several losses such as embedding loss, recovery loss and steganalysis loss. In previous research works, fixed loss weights are usually chosen for training optimization, and this setting is not linked to the importance of the steganography task itself and the training process. In this paper, we propose a Two-stage Curriculum Learning loss scheduler (TSCL) for balancing multinomial losses in deep learning image steganography algorithms. TSCL consists of two phases: a priori curriculum control and loss dynamics control. The first phase firstly focuses the model on learning the information embedding of the original image by controlling the loss weights in the multi-party adversarial training; secondly, it makes the model shift its learning focus to improving the decoding accuracy; and finally, it makes the model learn to generate a steganographic image that is resistant to steganalysis. In the second stage, the learning speed of each training task is evaluated by calculating the loss drop of the before and after iteration rounds to balance the learning of each task. Experimental results on three large public datasets, ALASKA2, VOC2012 and ImageNet, show that the proposed TSCL strategy improves the quality of steganography, decoding accuracy and security.

Figures (4)

• Figure 1: Discrete loss scheduling and continuous loss scheduling.
• Figure 2: Comparison between TSCL scheme and steganography images using only curriculum control and only loss control model.
• Figure 3: Comparison of cover image and stego image of TSCL scheme under 1-3 bpp capacity steganogayphy.
• Figure 4: Comparison of image steganography quality between TSCL scheme and baseline scheme model.