DeepSteg: Integerating new paradigms of cascaded deep video steganography for securing digital data
In today’s world, securing digital data and ensuring its confidentiality is crucial. Video steganography serves a critical role in this purpose, as it provides higher embedding capacity compared to images. Also, the complex structure of video files is designed to be robust against various attacks an...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-03-01
|
| Series: | Alexandria Engineering Journal |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S111001682401620X |
| _version_ | 1826898707219480576 |
|---|---|
| author | Sahar Magdy Sherin Youssef Karma M. Fathalla Saleh ElShehaby |
| author_facet | Sahar Magdy Sherin Youssef Karma M. Fathalla Saleh ElShehaby |
| author_sort | Sahar Magdy |
| collection | DOAJ |
| description | In today’s world, securing digital data and ensuring its confidentiality is crucial. Video steganography serves a critical role in this purpose, as it provides higher embedding capacity compared to images. Also, the complex structure of video files is designed to be robust against various attacks and distortions which adds an extra layer of security. In this study, a cascaded deep video steganography framework is proposed, integrating CryptoSteganography and deep video detection and tracking, where data is embedded at numerous tracked objects so the secret data location will be different from frame to frame ensuring the security of the data. The proposed model comprises three phases, detection of multi-objects, tracking the detected objects and using steganography to embed the data into the tracked objects. An enhanced detection architecture is proposed that captures the features from the 2D camera images and the LiDAR points through Deep leaning aggregation (DLA34) architecture before passing it through the Pyramid Split Architecture (PSA) to fuse the features. The detected objects are transferred to a novel tracking phase named E_SS architecture. To ensure the efficiency of the tracking module, an improved efficientNetV2 architecture named DAR_EfficientNetV2 was based on the DARNet Attention module. Finally, the tracked objects are cropped from the frame, where the steganography technique which is based on Discrete wavelet transform (DWT) and Singular Value Decomposition (SVD) is employed to embed the secret data whether it is images or text. The detection phase achieved an improvement of 3% in detected vehicles and trucks in the Nuscenes dataset, while motorcycle object detection improved by 9.5% in terms of mAP. Moreover, the E_SS tracking technique achieved an improvement rate of 2.1%.in tracking cars in the KITTI Dataset. Additionally, E_SS achieved an improvement rate of 7% than the previous state-of-arts. Furthermore, the deep Stegnagraphy technique was evaluated on multiple datasets ensuring its imperceptibility, high hiding capacity, and robustness against noise. An average PSNR of 73.57 was achieved along with an average PSNR of 70.84 achieved under noise attack. |
| first_indexed | 2025-02-17T06:39:29Z |
| format | Article |
| id | doaj.art-ac24519a20f5438697078317080b34bf |
| institution | Directory Open Access Journal |
| issn | 1110-0168 |
| language | English |
| last_indexed | 2025-02-17T06:39:29Z |
| publishDate | 2025-03-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Alexandria Engineering Journal |
| spelling | doaj.art-ac24519a20f5438697078317080b34bf2025-01-07T04:17:11ZengElsevierAlexandria Engineering Journal1110-01682025-03-01116483501DeepSteg: Integerating new paradigms of cascaded deep video steganography for securing digital dataSahar Magdy0Sherin Youssef1Karma M. Fathalla2Saleh ElShehaby3Computer Engineering Department, Pharos University, Alexandria, Egypt; Corresponding author.Computer Engineering Department, Arab Academy for Science and Technology, Alexandria, EgyptComputer Engineering Department, Arab Academy for Science and Technology, Alexandria, EgyptDepartment of Biomedical Engineering, Alexandria University, EgyptIn today’s world, securing digital data and ensuring its confidentiality is crucial. Video steganography serves a critical role in this purpose, as it provides higher embedding capacity compared to images. Also, the complex structure of video files is designed to be robust against various attacks and distortions which adds an extra layer of security. In this study, a cascaded deep video steganography framework is proposed, integrating CryptoSteganography and deep video detection and tracking, where data is embedded at numerous tracked objects so the secret data location will be different from frame to frame ensuring the security of the data. The proposed model comprises three phases, detection of multi-objects, tracking the detected objects and using steganography to embed the data into the tracked objects. An enhanced detection architecture is proposed that captures the features from the 2D camera images and the LiDAR points through Deep leaning aggregation (DLA34) architecture before passing it through the Pyramid Split Architecture (PSA) to fuse the features. The detected objects are transferred to a novel tracking phase named E_SS architecture. To ensure the efficiency of the tracking module, an improved efficientNetV2 architecture named DAR_EfficientNetV2 was based on the DARNet Attention module. Finally, the tracked objects are cropped from the frame, where the steganography technique which is based on Discrete wavelet transform (DWT) and Singular Value Decomposition (SVD) is employed to embed the secret data whether it is images or text. The detection phase achieved an improvement of 3% in detected vehicles and trucks in the Nuscenes dataset, while motorcycle object detection improved by 9.5% in terms of mAP. Moreover, the E_SS tracking technique achieved an improvement rate of 2.1%.in tracking cars in the KITTI Dataset. Additionally, E_SS achieved an improvement rate of 7% than the previous state-of-arts. Furthermore, the deep Stegnagraphy technique was evaluated on multiple datasets ensuring its imperceptibility, high hiding capacity, and robustness against noise. An average PSNR of 73.57 was achieved along with an average PSNR of 70.84 achieved under noise attack.http://www.sciencedirect.com/science/article/pii/S111001682401620XDeepLearningMulti-object trackingMulti-object detectionSteganographyCryptographyMedical data |
| spellingShingle | Sahar Magdy Sherin Youssef Karma M. Fathalla Saleh ElShehaby DeepSteg: Integerating new paradigms of cascaded deep video steganography for securing digital data Alexandria Engineering Journal DeepLearning Multi-object tracking Multi-object detection Steganography Cryptography Medical data |
| title | DeepSteg: Integerating new paradigms of cascaded deep video steganography for securing digital data |
| title_full | DeepSteg: Integerating new paradigms of cascaded deep video steganography for securing digital data |
| title_fullStr | DeepSteg: Integerating new paradigms of cascaded deep video steganography for securing digital data |
| title_full_unstemmed | DeepSteg: Integerating new paradigms of cascaded deep video steganography for securing digital data |
| title_short | DeepSteg: Integerating new paradigms of cascaded deep video steganography for securing digital data |
| title_sort | deepsteg integerating new paradigms of cascaded deep video steganography for securing digital data |
| topic | DeepLearning Multi-object tracking Multi-object detection Steganography Cryptography Medical data |
| url | http://www.sciencedirect.com/science/article/pii/S111001682401620X |
| work_keys_str_mv | AT saharmagdy deepstegintegeratingnewparadigmsofcascadeddeepvideosteganographyforsecuringdigitaldata AT sherinyoussef deepstegintegeratingnewparadigmsofcascadeddeepvideosteganographyforsecuringdigitaldata AT karmamfathalla deepstegintegeratingnewparadigmsofcascadeddeepvideosteganographyforsecuringdigitaldata AT salehelshehaby deepstegintegeratingnewparadigmsofcascadeddeepvideosteganographyforsecuringdigitaldata |