Secure Visual Data Encryption in Cloud Computing Using Hybrid CNN-LSTM and Frequency Domain Transformations

Secure visual data protection has become a critical requirement in cloud computing environments due to the increasing volume of sensitive images stored and transmitted over untrusted infrastructures. Hybrid encryption schemes that combine deep learning and frequency-domain transformations offer promising solutions by enhancing both confusion and diffusion characteristics. Convolutional Neural Networks (CNN) are effective in capturing spatial features, while Long Short-Term Memory (LSTM) networks model longrange dependencies, making their integration suitable for complex image encryption tasks. In this work, a hybrid visual encryption framework is developed by combining CNN–LSTMbased encryption with Discrete Cosine Transform (DCT), zigzag scanning, XOR-based secret key encryption, and visual cryptography. The proposed scheme performs client-side encryption to ensure data confidentiality in cloud storage scenarios and prevents unauthorized reconstruction through share-based protection. Experimental evaluation is conducted using Peak Signal-to-Noise Ratio (PSNR) and Mean Squared Error (MSE) to assess reconstruction quality after decryption. The results demonstrate that the proposed approach achieves reliable image recovery with minimal perceptual degradation while providing enhanced security and computational efficiency suitable for cloud-based visual data applications.