Abstract: In the field of visual camouflage, generating a high-quality background image that seamlessly blends with complex foreground objects and diverse background environments is a critical task. When dealing with such complex scenes, the existing techniques have insufficient foreground feature extraction, resulting in insufficient fusion of the generated background image with the foreground objects, making it difficult to achieve the desired camouflage effect. In order to solve this problem and achieve the goal of higher quality visual camouflage effect, this paper proposes a new foreground feature-guided camouflage image generation method (Object Enhancement Module - Diffusion Refinement , OEM-DR), which generates camouflage images by enhancing the foreground features to guide the background. The method firstly designs a new object enhancement module to optimize the attention mechanism of the model, and eliminates the attention weights that have less influence on the output through pruning strategy, so that the model focuses more on the key features of the foreground objects, and thus guides the generation of the background more effectively. Second, a novel detail optimization framework based on diffusion strategy is constructed, which maintains the integrity of the global structure of the image while performing fine optimization processing on the local details of the image. In experiments on standard camouflaged image datasets, the proposed method in this study achieves significant improvement in both FID (Fréchet Inception Distance) and KID (Kullback-Leibler Divergence) evaluation metrics, which verifies the feasibility of the method. This suggests that by strengthening foreground features and detail optimization, the fusion between background images and foreground objects can be effectively improved to achieve higher quality visual camouflage effects.

Yuelin Chen, Yuefan An, Yonsen Huang and Xiaodong Cai, “Foreground Feature-Guided Camouflage Image Generation” International Journal of Advanced Computer Science and Applications(IJACSA), 16(1), 2025. http://dx.doi.org/10.14569/IJACSA.2025.0160140

@article{Chen2025,
title = {Foreground Feature-Guided Camouflage Image Generation},
journal = {International Journal of Advanced Computer Science and Applications},
doi = {10.14569/IJACSA.2025.0160140},
url = {http://dx.doi.org/10.14569/IJACSA.2025.0160140},
year = {2025},
publisher = {The Science and Information Organization},
volume = {16},
number = {1},
author = {Yuelin Chen and Yuefan An and Yonsen Huang and Xiaodong Cai}
}