Hybrid Image Compression-Encryption Scheme with Chaotic Logistic Map and XOR with Variational Autoencoder

With the increase in the speed of digital data production and transmission, protecting sensitive content and preventing unauthorized access has become a critical requirement. In this context, data security and encryption methods are indispensable components of modern information systems. Meanwhile, the increasing volume of data necessitates transmission at lower bandwidth and storage costs. This situation demands the use of effective compression techniques to enhance efficiency. In this study, a hybrid system is proposed for both compression and secure transmission of visual data. Based on the Variational Autoencoder (VAE) model, exclusive OR (XOR) and chaotic system-based encryption methods were integrated. Three model-method combinations (VAE, VAE + XOR, and VAE + Chaotic) were implemented. The performance of these model methods was comprehensively analyzed using multi-faceted metrics, including visual quality (PSNR, SSIM), security (NPCR, UACI), statistical analysis (entropy, correlation, histogram), processing time (encoding/decoding, encryption/decryption), compression ratio, and hardware resource usage (CPU, RAM, GPU). Experimental studies were conducted using the FEI face image dataset. According to the results, the VAE + Chaotic model method stands out as the most successful and balanced solution, offering high reconstruction quality, strong security features, low hardware usage, and fast processing time. Overall, this study demonstrates that deep learning-based models provide an effective alternative for secure visual data transmission, especially in resource-constrained environments.