Authors - Anamika Saini, Kavita Rathi Abstract - Video steganography has emerged as an effective approach for secure multimedia communication by concealing secret information inside video frames while maintaining visual imperceptibility. This work presents a UCF101 datasetbased video steganography framework using a 2LSB embedding technique for hiding secret image data inside video frames. 101 video samples from the UCF101 benchmark dataset were utilized to evaluate the robustness and scalability of the proposed framework under diverse motion and background conditions. The visual quality of stego frames was analyzed using Peak Signal-to- Noise Ratio (PSNR), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). In addition, machine learning-based steganalysis models including Support Vector Machine (SVM), Convolutional Neural Network (CNN), and Extreme Gradient Boosting (XGBoost) were implemented to detect hidden information from cover and stego frames. Experimental results demonstrate that the proposed embedding method maintains high visual quality with low distortion values. However, the steganalysis results indicate that advanced machine learning approaches, particularly XGBoost, can effectively identify hidden embedding patterns present in stego frames. The study highlights the trade-off between visual imperceptibility and resistance against machine learning-based steganalysis in modern video steganography systems.
Open Zoom