Article: Adaptive highly secure and imperceptible video steganography for medical images via deep convolutional neural network Journal: International Journal of Information and Computer Security (IJICS) 2024 Vol.25 No.1/2 pp.78 - 95 Abstract: The risk of medical data theft has increased due to technological developments in communication. Steganography is frequently used for covert communication; however, existing steganography algorithms have poor imperceptibility, lesser embedding capacity, and lower extraction accuracy. The proposed model extracts the keyframe followed by data hiding and extraction using a convolutional neural network (CNN). The CNN model incorporates preparation, hidden and reveal networks that extract the low and high-dimensional features from the keyframe and secret image via convolution, and pooling operations. All three networks are trained collectively to minimise the loss function value. The computed value of peak signal to noise ratio (PSNR) between the keyframe and container frame is 32.7 decibels (dB), whereas an embedding capacity of 100% demonstrates the higher imperceptibility and embedding capacity of the proposed model than other state of art methods. The normalize correlation value (NC) of 0.996 on the receiver side indicates better robustness against attacks. Inderscience Publishers - linking academia, business and industry through research

Title: Adaptive highly secure and imperceptible video steganography for medical images via deep convolutional neural network

Authors: Sahil Gupta; Naresh Kumar Garg

Addresses: Department of Electronics and Communication Engineering, Giani Zail Singh Campus College of Engineering and Technology, Maharaja Ranjit Singh Punjab Technical University, Bathinda, Punjab, India ' Department of Computer Science and Engineering, Giani Zail Singh Campus College of Engineering and Technology, Maharaja Ranjit Singh Punjab Technical University, Bathinda, Punjab, India

Abstract: The risk of medical data theft has increased due to technological developments in communication. Steganography is frequently used for covert communication; however, existing steganography algorithms have poor imperceptibility, lesser embedding capacity, and lower extraction accuracy. The proposed model extracts the keyframe followed by data hiding and extraction using a convolutional neural network (CNN). The CNN model incorporates preparation, hidden and reveal networks that extract the low and high-dimensional features from the keyframe and secret image via convolution, and pooling operations. All three networks are trained collectively to minimise the loss function value. The computed value of peak signal to noise ratio (PSNR) between the keyframe and container frame is 32.7 decibels (dB), whereas an embedding capacity of 100% demonstrates the higher imperceptibility and embedding capacity of the proposed model than other state of art methods. The normalize correlation value (NC) of 0.996 on the receiver side indicates better robustness against attacks.

Keywords: convolutional neural network; CNN; imperceptibility; steganography; security; peak signal to noise ratio; PSNR; structural similarity index; SSIM.

DOI: 10.1504/IJICS.2024.142702

International Journal of Information and Computer Security, 2024 Vol.25 No.1/2, pp.78 - 95

Received: 18 Jun 2023
Accepted: 17 Sep 2023
Published online: 18 Nov 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Adaptive highly secure and imperceptible video steganography for medical images via deep convolutional neural network

Keep up-to-date