Reinforced source camera identification using non-decimated wavelet transform

The Sensor Pattern Noise (SPN) extracted from digital pictures can be interpreted as a unique sensor fingerprint for a digital camera and can be used to perform source identification of digital cameras. Scene details can contaminate SPN signatures. This paper presents a method to extract the SPN by applying non-decimated wavelet transform to digital pictures and then disinfect the contaminated SPN in order to improve the identification rate of the SPNs. The coefficients within the resulting wavelet high frequency sub-bands are filtered to extract the SPN of the image. By using non-decimated wavelet transform, we perform a two-step comparison technique that first isolates all the contaminated components of the SPN and neutralise these components from a contaminated SPN. The reinforced SPN is then matched against the corresponding components in the reference camera fingerprint. The two-step comparison technique provides a reinforced SPN of reduced contamination for the matching against the camera reference fingerprint. Experimental results were performed using images of ten cameras to identify the source camera of the images. Results show that the proposed technique generates superior results to that of the non-reinforced SPNs.

http://icbisp2017.events.theiet.org.cn/