Recent years, with the rapid development of information technology, the information security has received more and more attention. A variety of encryption methods to protect the information have been reported. Visual cryptography is one of the encryption methods, which has highly security because of its threshold feature. And the cryptographic information can be explained by a naked eye in the decryption process. In the application of visual cryptography, however, each shared image is limited to transparency films and overlapping on computer. In our previous work, we proposed the scheme of invisible visual cryptography and developed the visual-cryptography-based optical hiding system (VCOH), which transformed the conventional visual cryptography shares into diffraction optical elements (DOEs). It not only increases the application range of visual cryptography, but also enhances security. In this paper, we propose an optical information hiding system based on the extended visual cryptography, which inherits the concept of invisible visual cryptography. In contrast to our previous work, the method proposed in this work can hide a meaningful image instead of text messages. Meanwhile, the capacity and imperceptibility of the method are greatly increased. The hiding process of the system contains two steps. Firstly, the secret image is converted into meaningful shares through the extended visual cryptography algorithm. Secondly, the meaningful shares are able to hide in phase-keys through an iterative phase retrieval algorithm, such as Gerchberg-Saxton algorithm and Yang-Gu iterative algorithm. Then the phase-keys can be made into diffraction optical elements (DOEs) to store and transport in a physical way. In the decryption process, DOEs are illuminated with the laser beam to reconstruct the meaningful shares. The secret image can be explained by the direct overlapping of the reconstructed shares without any optical or cryptographic knowledge. The simulation and optical experimental results show that the proposed method has good performance of security and validate the feasibility of the proposed method. Besides, in this paper the robustness and security issues are also analyzed. This system has a high security because of its indistinguishability under adaptive chosen ciphertext attack (IND-CCA2) security. Additionally, this system is relatively less robust than the VCOH because it shares meaningful images with highly complex and detailed structures.