Source Camera Identification Using GGD and Normalized DCT Model-Based Feature Extraction

Abstract

In recent decades, blind feature-based source camera detection has drawn a lot of attention. In literature, researchers used a specific sort of distortion, such as vignetting effects, chromatic aberration, and radial lens distortion, etc., to distinguish between different camera models. Therefore, it becomes specific to the distortions present in the particular camera model. However, distortion-specific approaches perform poorly in the absence of the specific distortion in an image. To develop a source camera identification system, we introduce a non-distortion-specific methodology using normalized DCT coefficients, a generalized Gaussian distribution (GGD) model-based blind features. DCT is performed on the sub-images of N × N blocks after the image is normalized using mean subtracted contrast normalization (MSCN). Over three scales, the DCT features are extracted. To perform camera model identification, a multi-class SVM is used to utilize all the extracted features. The experiment was conducted on the Dresden image database, which contains 10,173 images captured by 43 devices distributed over 12 camera models. The proposed method showed a maximum accuracy of 98.82% on the dataset.