Building fingerprints with information from three color bands for source camera identification

Sensor Pattern Noise Estimation Based on Improved Locally Adaptive DCT Filtering and Weighted Averaging for Source Camera Identification and Verification

“…The results shown in Table VII reveal an interesting finding in that the PRNU extracted from gray level images seems to offer better source camera identification than that of the green channel in both datasets. Also, it is clear that the combination of the three RGB channels as proposed in [22] outperforms the methods which use the green channel only, the Luminance image and the luminance PRNU. Finally, the proposed color concatenation achieves the best performance among the tested methods.…”

“…On the other hand, if the green channel only is used, the physical PRNU information which could exist in other color components (red and blue) is not taken into account. In [22], a non-linear combination is applied by extracting the PRNU from each color channel separately, and then the largest coefficient in magnitude at each location is chosen. In this work, the PRNU is estimated from each channel separately and then the resulting PRNUs are concatenated to form a color PRNU.…”

“…That is, the green channel-based PRNU as suggested in [11] [27], the luminance image-based PRNU estimation as mentioned in [19] [25], the luminance PRNU method as proposed in [8] [17] [26], and finally the color combination scheme developed in [22]. Note that the authors in [22] developed three schemes but scheme 3 is used here because it has been found to deliver the best performance. The results shown in Table VII reveal an interesting finding in that the PRNU extracted from gray level images seems to offer better source camera identification than that of the green channel in both datasets.…”

“…Indeed, the authors in [21] proposed a color decoupling algorithm prior to the filtering stage to reduce the color interpolation noise which is caused by the CFA. In [22], three color combination schemes were proposed to obtain the final PRNU using the red, green and blue channels. The basic idea is to extract the PRNU from each color channel separately and then select the pixel with the largest magnitude.…”