Efficient steganalysis using convolutional auto encoder network to ensure original image quality.

Area of Science:

• Computer Science
• Digital Image Processing
• Cybersecurity

Background:

• Steganalysis, the detection of hidden information in images, is crucial but computationally challenging.
• Existing deep learning methods for steganalysis often neglect image noise, increasing computational overhead and reducing accuracy.
• Noise in images can lead to inaccurate predictions in steganalysis classification techniques.

Purpose of the Study:

• To develop an efficient steganalysis technique that addresses the limitations of existing methods, particularly concerning image noise.
• To improve the accuracy and reduce the computational overhead of predicting hidden information in images.
• To introduce a novel deep learning model capable of handling non-Gaussian noise in image steganalysis.

Main Methods:

• A new method, Non-Gaussian Noise Aware Auto Encoder Convolutional Neural Network (NGN-AEDNN), was proposed.
• Non-Gaussian noise removal techniques were applied before the learning task to enhance accuracy.
• Gaussian noise removal was integrated into each neural network iteration to manage error rates without noisy features.

Main Results:

• The proposed NGN-AEDNN technique demonstrated efficient steganalysis.
• The method achieved accurate learning tasks by effectively removing image noise.
• Simulations in Matlab confirmed reduced computational overhead compared to existing methods.

Conclusions:

• The NGN-AEDNN model provides an efficient solution for image steganalysis.
• The noise removal techniques integrated into the model significantly improve learning accuracy and reduce computational load.
• This approach offers a more robust and computationally efficient method for detecting hidden information in images.