Multiple Description Coding for Encrypted Images

This work proposes a novel Multiple Description Coding (MDC) method for encrypted images transmitted over unreliable networks. Here, a pseudorandom permutation is used to mask the original content to protect privacy. The encoder splits the encrypted data into two descriptions, and each description includes different rigid pixels and coefficients generated from orthogonal transform to the elastic pixels. An acceptable quality reconstruction can be decodable from the side decoder through iteratively updating the values of coefficients based on the spatial correlation. At the central decoder, since the more accurate rigid pixels received, the more similar predicted image can be considered as initialization. Moreover, the operation of iterative reconstruction twice to unknown pixels also makes the estimations closer to the real values. Therefore, a high quality reconstruction from multiple descriptions can be obtained at the central decoder. Comparison is made with the single description coding and the experiments result is shown that the reconstruction performance with the multiple description coding for encrypted images is significantly improved.