Analysis of Video Stream for Known Plaintext Attack and Perceptual Encryption Feature

Abstract

  Secure transmission of digital video must Coordinate two competing goals: protecting the content against cryptographic attacks while preserving format compliance and low complexity for real-time multimedia delivery. This paper presents an integrated joint video coding-and-encryption framework in which a convolutional neural network (CNN) cipher is embedded into a modified MPEG-2 codec. The cipher operates as a keyed value transformation whose weights and biases are generated from a chaotic logistic map, with the control parameter μ and the initial value x₀ acting as the secret key encryption is applied in the coded domain to the quantized transform coefficients and motion vectors before entropy coding. Two operating modes are supported: a full-protection mode that encrypts coefficients and motion vectors, and a perceptual mode that encrypts only the motion-vector stream together with a reduced quality factor, yielding a controllable low-quality preview suitable for try-before-buy services. Building on a focused review of prior work on known-plaintext attacks (KPA) and perceptual encryption, the scheme is assessed against the KPA threat through an XOR-based analysis, which shows that the effective mask recovered from one plaintext–ciphertext pair cannot decrypt a different encrypted scene and instead produces a noise-like result. Experimental results demonstrate that the proposed framework resists the known-plaintext attack while providing controllable perceptual protection, confirming its suitability for secure and efficient video transmission.