Performance Evaluation And Deep Learning Steganalysis Of Covert APK Transmission Using Video Steganography Using 2D Haar Discrete Wavelet Transform And LSB Substitution
DOI:
https://doi.org/10.70917/ijcisim-2026-2285Keywords:
video steganography, APK binary embedding, 2D Haar DWT, LSB substitution, Hamming (7,4) error correction, XOR encryption, Steganalysis, VGG16 fine-tuning, CNN-LSTM, NPCR, UACI, Shannon entropy, Android security.Abstract
The hidden payload recovery should be exact rather than approximated as there is zero tolerance for byte corruption in APK files. WE aim to develop a system which ensure zero-byte-error recovery for the APK while keeping the stego-video visually indistinguishable from a normal video. We have developed a complete pipeline for covert Android Package file transmission through video steganography, addressing the one engineering requirement that distinguishes APK embedding from every prior payload class in the field: Android's PackageManager rejects any recovered binary containing a single incorrect byte, making zero-error recovery a hard constraint rather than a quality target. The research demonstrates that Hamming (7,4) forward error correction at code rate 4/7 successfully satisfies this constraint across the embedding-extraction cycle, protecting the APK byte stream before LH sub-band coefficient modification and reconstructing it without residual errors under lossless storage conditions. The implementation of 2D Haar Discrete Wavelet Transform embedding in the horizontal-detail LH sub-band proved effective at maintaining imperceptibility despite the elevated payload densities that APK file sizes require, achieving PSNR of 47.3 to 52.1 dB, SSIM above 0.99, NPCR of 89.66%, and UACI of 0.82% within the accepted imperceptibility range. The implementation of VGG16 fine-tuning for spatial LH artefact detection proved effective at classifying stego frames with 91.52% accuracy and F1-score of 91.42%, successfully distinguishing embedded from clean frames despite the absence of prior training data for this specific payload class. The results have validated the effectiveness of frequency-domain binary payload embedding and highlight the potential for neural steganalysis as a practical detection mechanism for APK-in-video covert channels. The integration of all three pipelines into a single deployable desktop GUI application represents an advance not present in any prior covert APK transmission system.