Security of feature extraction in image hashing

Security and robustness are two important requirements for image hash functions. In this paper, we introduce "differential entropy" as a metric to quantify the amount of randomness in image hash functions and to study their security. We present a mathematical framework and derive expressions for the proposed security metric for various common image hashing schemes. Using the proposed security metric, we discuss the trade-offs between the security and robustness in image hashing.