Extreme learning machine based optimal embedding location finder for image steganography.

Area of Science:

• Computer Science
• Information Security
• Digital Image Processing

Background:

• Image steganography faces challenges in selecting optimal embedding locations to minimize host medium distortion.
• Existing methods struggle to achieve least deformation for secret message embedding.

Purpose of the Study:

• To propose a novel, high-performance approach for image steganography.
• To develop a supervised mathematical model using extreme learning machine (ELM) for optimal embedding location selection.

Main Methods:

• Modified extreme learning machine (ELM) trained on image texture features (contrast, homogeneity).
• ELM utilized in regression mode to predict optimal embedding locations and evaluation metrics.
• Techniques to counteract overfitting during ELM training.

Main Results:

• The modified ELM approach outperforms existing methods in imperceptibility.
• Achieved up to 28% improvement in imperceptibility compared to state-of-the-art methods.
• Evaluation metrics include correlation, Structural Similarity Index (SSIM), fusion matrices, and Mean Square Error (MSE).

Conclusions:

• The proposed steganography approach is highly proficient in preserving visual image information.
• The modified ELM effectively identifies optimal embedding locations with minimal distortion.
• This method offers significant advancements in steganographic imperceptibility.