With the development of data security and privacy requirements in the field of cloud computing, Reversible Data Hiding in Encrypted Images (RDHEI) in encryption domain has received increasing attention. In order to take full advantage of the spatial and textural features of the original image, reversible data hiding in encrypted image with adaptive Huffman code based on Dynamic Prediction Axes (RDHEI-HDA) is proposed. First, the prediction errors of the original plaintext image are calculated according to the multidirectional median edge detector (M-MED) combined with the Dynamic Prediction Axes which are generated by the spatial correlation of the original image. After encryption process with the stream cipher, the adaptive Huffman coding labeling rule is created for pixel labeling and classification according to the Dynamic Prediction Axes and the distribution of prediction errors. Finally, bit substitution is employed to insert secret data and side information into the image. In Comparison to most of the state-of-the-art RDHEI methods, the experimental results show that the RDHEI-HDA method provides a higher pure payload while ensuring safety.