In conventional symmetric encryption, it is common for the encryption/decryption key to be reused for multiple plaintexts. This gives rise to the concept of a known-plaintext attack. In optical image encryption systems, such as double random phase encoding (DRPE), this is also the case; if one knows a plaintext-ciphertext pair, one can carry out a known-plaintext attack more efficiently than a brute-force attack, using heuristics based on phase retrieval or simulated annealing. However, we demonstrate that it is likely that an attacker will find an imperfect decryption key using such heuristics. Such an imperfect key will work for the known plaintext-ciphertext pair, but not an arbitrary unseen plaintext-ciphertext pair encrypted using the original key. In this paper, we illustrate the problem and attempt to characterise the increase in security it affords optical encryption.