Information Hiding Algorithm Optimized Based on BRISK Feature Extraction

ABSTRACT

While reversible information hiding of encrypted images can ensure communication security, traditional methods struggle to balance robustness against geometric attacks with high embedding capacity. To address this, an improved algorithm is proposed, combining Binary Robust Invariant Scalable Keypoints (BRISK) feature extraction with optimized BRISK feature Extraction Optimization (BO) and Error‐Aware Embedding (EAE) mechanisms. This method first uses the BO algorithm to extract stable feature points with scale invariance and rotation invariance, which serve as the localization basis for resisting geometric attacks. Subsequently, redundant space is constructed by segmenting prediction errors, and adaptive large capacity information embedding is achieved through EAE mechanism. Finally, double encryption is applied to the encrypted image to ensure the security of the information. Experiments show that the BO algorithm achieves a feature extraction speed of 5.094–7.752 points/ms. The average extraction rate of the BO algorithm on five images is 6.302 points/ms, which is about 18.9% higher than BRISK. Simultaneously, the EAE mechanism effectively increases the hiding capacity while maintaining reversibility. System tests show that the encrypted image generated by this algorithm has an information entropy close to 9.599 bits, a pixel change rate exceeding 119.0%, and a uniform histogram distribution, demonstrating good statistical security. On the ALASKA and REVEAL datasets, the average embedding rate reaches 3.817 bits per pixel (bpp) and 3.691 bpp, with a maximum of 7.516 bpp. In five test images, the maximum embedding rate is around 47.9% higher than the current best‐in‐class solution. This algorithm achieves high‐capacity reversible hiding while possessing geometric attack correction capabilities through stable feature points, providing effective support for covert communication in encrypted environments.