Reversible Data Hiding By Using CNN Prediction and Adaptive Embedding

Area of Science:

• Computer Science
• Information Security
• Digital Image Processing

Background:

• Reversible data hiding (RDH) is crucial for embedding information in digital images with minimal distortion.
• Existing RDH methods face challenges in accurately predicting image content and minimizing embedding-induced visual artifacts.

Purpose of the Study:

• To propose a novel and efficient RDH method that enhances prediction accuracy and reduces embedding distortion.
• To improve the visual quality of steganographic images while maintaining high embedding capacity.

Main Methods:

• Developed a convolutional neural network (CNN) based predictor that utilizes image partitioning and contextual information from multiple parts for improved prediction.
• Implemented a prediction-error-ordering (PEO) based adaptive embedding strategy that leverages background complexity to select and pair prediction errors for data hiding.
• Employed multi-receptive fields and global optimization in the CNN predictor to leverage more neighboring pixels for enhanced prediction performance.

Main Results:

• Achieved a high average Peak Signal-to-Noise Ratio (PSNR) of 63.59 dB on the Kodak dataset for an embedding capacity of 10,000 bits.
• Demonstrated superior performance compared to existing state-of-the-art RDH methods in terms of visual quality and embedding efficiency.
• The proposed method effectively reduces embedding distortion by adaptively embedding data based on prediction errors and image content complexity.

Conclusions:

• The proposed CNN-based predictor and PEO adaptive embedding strategy offer a significant advancement in reversible data hiding.
• The method provides excellent visual quality and high embedding capacity, making it suitable for secure steganography applications.
• This research contributes to the development of more efficient and visually imperceptible data hiding techniques.