Open Access

ARTICLE

VARStego: A Reversible Local Variance-Based Steganographic Method

Basten Andika Salim¹, Adifa Widyadhani Chanda D’Layla¹, Ntivuguruzwa Jean De La Croix², Tohari Ahmad^1,*, Kambombo Mtonga³

1 Department of Informatics, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia
2 Faculty of Computing and Information Sciences, University of Lay Adventists of Kigali (UNILAK), Kigali, Rwanda
3 School of Science and Technology, Malawi University of Business and Applied Sciences, Blantyre, Malawi

* Corresponding Author: Tohari Ahmad. Email:

Computers, Materials & Continua 2026, 89(2), 60 https://doi.org/10.32604/cmc.2026.085342

Received 09 May 2026; Accepted 29 July 2026; Issue published 15 September 2026

Abstract

The modernization of communication and healthcare environments has introduced critical security challenges, as all transmission is almost certainly done through digital networks prone to attacks. To counteract this, researchers have worked to create methods of data concealment, hiding the existence of sensitive data itself from prying eyes. However, many of these methods lack the necessary ability to balance imperceptibility and payload capacity. In addition, different from generic images, medical images must preserve their structural and visual integrity, requiring frameworks that prioritize maintaining high similarity between images or, at times, complete recovery of the original image. This paper introduces VARStego, a reversible dual-layered steganographic framework designed to conceal confidential patient information within medical images. VARStego uses the Huffman algorithm to compress all patient information in a separate layer of the framework, while a localized variance-based algorithm is employed to analyze medical images to find rough and complex regions. Utilizing these regions, the VARStego embeds data within selected pixels using a redundancy-based algorithm, further improving payload capacity. The proposed VARStego method is tested on the DICOM Library dataset, containing Digital Imaging and Communications in Medicine (DICOM) formatted files and the Kaggle dataset, providing computed tomography (CT) medical images. VARStego achieved a Peak Signal-to-Noise Ratio (PSNR) of 72.192 dB and a perfect Structural Similarity Index Measure of up to 1, ensuring that image quality is maintained. These scores show minimal degradation as the payload size increases from 1 to 50 kilobyte for American Standard Code for Information Interchange (ASCII) payload and 1 to 100 kilobit for bitstring payload, demonstrating the method’s consistent performance.

---

Keywords

---