Using image sensor PUF as root of trust for birthmarking of perceptual image hash

The perceptual image hash has been widely used for integrity check of digital image content. Existing image hashes fail to identify the origin of image acquisition and non-repudiable authentication is contingent on the privacy of secret key. This paper presents a new and more secure image hashing scheme by exploiting the CMOS image sensor physical unclonable function (PUF) as a root of trust to imprint a birthmark into the image hash. The hash vector is generated directly by the CMOS image sensor based on the content-based features extracted from its captured image and the timestamp of the image without the need for a shared secret key for its authentication. Other than the ability to detect malicious tampering of image content from normal benign image processing operations, the use of CMOS image sensor PUF as a random one-way function to map the extracted feature and the timestamp to a hash vector provides a non-repudiable proof of the original image acquisition device and time of offence when the digital image content is to be presented as an evidence of a crime. As the hash vector can only be generated by the image sensor when it is powered on, it is inherently secure against invasive, data remanence and replay attacks, which are common threats to key-based perceptual image hashes. Our preliminary experimental results on 49 content-preserving and 5 wilfully tampered copies of an original image captured by a 64 × 64 image sensor PUF fabricated in 180 nm 3.3 V CMOS technology have demonstrated that the tampered images and their regions of alteration can be successfully detected from the hash vectors.