Digital Videos (DVs) are a powerful mean of communication and they are spreading on Social Networks (SNs) more and more every day (i.e. ~ 100 million of hours of video viewed per day1 in 2016). H.264 AVC is a common video format used to encode videos. It has been proved to be very effective for HD videos, but when the video resolution increases (4K and 8K) the efficiency decreases. For this reason in the next years more and more Online Social Networks (OSNs) will move to the new H.265 HEVC standard. It is widely known that Pixel Non Uniformity (PNU) noise can be exploited to identify the camera that acquired an image applying the method introduced by Fridrich et al. The same technique has been adapted to identify video DVs source camera. Unfortunately the noise present in a digital image heavily depends on the compression level used to store it and high (lossy) compression algorithms, can completely filter out any kind of noise. In this paper we analyze the effects of H.265 encoding on PNU noise, to prove under which conditions it still can be exploited for Source Camera Identification (SCI) for DV. We present the results of several experiments explicitly designed to point out the limits of this technique when dealing with H.265 videos.
Effects of H.265 encoding on PNU and source camera identification
Bruno A.
Conceptualization
;Cattaneo G.Supervision
2018
Abstract
Digital Videos (DVs) are a powerful mean of communication and they are spreading on Social Networks (SNs) more and more every day (i.e. ~ 100 million of hours of video viewed per day1 in 2016). H.264 AVC is a common video format used to encode videos. It has been proved to be very effective for HD videos, but when the video resolution increases (4K and 8K) the efficiency decreases. For this reason in the next years more and more Online Social Networks (OSNs) will move to the new H.265 HEVC standard. It is widely known that Pixel Non Uniformity (PNU) noise can be exploited to identify the camera that acquired an image applying the method introduced by Fridrich et al. The same technique has been adapted to identify video DVs source camera. Unfortunately the noise present in a digital image heavily depends on the compression level used to store it and high (lossy) compression algorithms, can completely filter out any kind of noise. In this paper we analyze the effects of H.265 encoding on PNU noise, to prove under which conditions it still can be exploited for Source Camera Identification (SCI) for DV. We present the results of several experiments explicitly designed to point out the limits of this technique when dealing with H.265 videos.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.