Evaluation of Wavelets in Image Camera Source Identification

Application of wavelet transform for image camera source identification has been widely reported in the literature and the written techniques use different wavelets. Due to the wavelets’ diversity and properties, it is beneficial for the research community to identify the best-performing wavelets for this application. This paper presents results for assessing the performance of the conventional wavelet-based image camera source identification technique against forty-one wavelets from Daubechies, Biorthogonal, Symlets, and Coiflets wavelet families. VISION image dataset comprising 34,427 images captured by eleven camera brands of thirty-five models was used to generate experimental results. Hundred plane images from each camera brand dataset were randomly selected and used to generate experimental results, where 70% of each dataset’s images were used to compute the camera brand’s signature, and 30% of the images were used to assess the performance of the method. Normalized cross-correlation of the camera brand signature and calculated image noise were used to find the camera match. To compare the method’s performance when using different wavelets, a new assessment criterion was introduced and used to quantify the method’s performance across images of different camera brands. Results show that the conventional wavelet-based image camera source identification achieves its highest performance when it uses sym2 closely followed by coif1 wavelets.