Tiwari Mayank, Gupta Bhupendra
Indian Institute of Information Technology, Design & Manufacturing Jabalpur, MP 482005, India.
Indian Institute of Information Technology, Design & Manufacturing Jabalpur, MP 482005, India.
Forensic Sci Int. 2018 Apr;285:111-120. doi: 10.1016/j.forsciint.2018.02.005. Epub 2018 Feb 15.
For source camera identification (SCI), photo response non-uniformity (PRNU) has been widely used as the fingerprint of the camera. The PRNU is extracted from the image by applying a de-noising filter then taking the difference between the original image and the de-noised image. However, it is observed that intensity-based features and high-frequency details (edges and texture) of the image, effect quality of the extracted PRNU. This effects correlation calculation and creates problems in SCI. For solving this problem, we propose a weighting function based on image features. We have experimentally identified image features (intensity and high-frequency contents) effect on the estimated PRNU, and then develop a weighting function which gives higher weights to image regions which give reliable PRNU and at the same point it gives comparatively less weights to the image regions which do not give reliable PRNU. Experimental results show that the proposed weighting function is able to improve the accuracy of SCI up to a great extent.
对于源相机识别（SCI），光响应非均匀性（PRNU）已被广泛用作相机的指纹特征。通过应用去噪滤波器从图像中提取PRNU，然后取原始图像与去噪图像之间的差值。然而，据观察，图像的基于强度的特征和高频细节（边缘和纹理）会影响提取的PRNU的质量。这会影响相关性计算并在SCI中产生问题。为了解决这个问题，我们提出了一种基于图像特征的加权函数。我们通过实验确定了图像特征（强度和高频内容）对估计的PRNU的影响，然后开发了一种加权函数，该函数对能给出可靠PRNU的图像区域赋予较高权重，同时对不能给出可靠PRNU的图像区域赋予相对较低的权重。实验结果表明，所提出的加权函数能够在很大程度上提高SCI的准确性。
