Homomorphic encryption, which protects privacy effectively and allows algebraic operations directly in the ciphertext, has been a active topic in the study of cloud computing. Due to security threats in cloud computing, the security protection and integrity authentication of encrypted data remain critical problems. The challenge lies in how to retrieve the encrypted data. To achieve more effective management and security protection of encrypted images on-line, this paper proposes a reversible data hiding scheme for ciphertext based on the public key cryptosystems with homomorphic and probabilistic properties. In the proposed scheme, partial pixels are selected as target pixels by a secret key and all bits of the target pixels are embedded into the other pixels with difference expansion (DE) to vacate room before encryption. As a bonus, secret data can be embedded directly in homomorphic encrypted domain by altering the target pixels with the fake pixels which are comprised of secret data. With the legal key, the receiver can extract the embedded data from the encrypted image and the directly decrypted image. Furthermore, user can accurately recover the original image after decryption and data extraction. Finally, experimental results show that extra data can be embedded more efficiently in homomorphic encrypted domain while keeping the quantity of data unchanged. Besides, the embedded data can be extracted in both ciphertext and plaintext.