An efficient multi-predictor reversible data hiding algorithm based on performance evaluation of different prediction schemes

With the broad development and evolution of digital data exchange, security has become an important issue in data storage and transmission since digital data can be easily manipulated and modified. Reversible data hiding algorithms are special class of steganography that are capable of recovering the original cover image upon the extraction of the secret data. This issue is of interest in medical and military imaging applications. Many algorithms in this class exploit the idea of prediction in order to increase the embedding capacity as well as the quality of the stego image. However, the performance of these algorithms depends on the type of predictor that is being used. The main goal in this paper is to survey different predictors and evaluate their performance when employed in two classical reversible data hiding algorithms. The evaluation considered plugging 22 predictors in the two algorithms to process 1438 test images. Experimental results validated the varying capabilities of different predictors and showed that the non-causal median predictor had the best performance in the two algorithms. Further more, the paper proposes a new multi-predictor reversible data hiding algorithm. Basically, the algorithm employs multiple predictors in an extended version of the modification of prediction errors (MPE) algorithm. The algorithm takes advantage of the results obtained from the performance evaluation of different predictors to select the best set of predictors. Performance evaluation proved the ability of the proposed algorithm in increasing the embedding capacity while maintaining high stego image quality.