On the performance of Li's unsupervised image classifier and the optimal cropping position of images for forensic investigations

Images from digital imaging devices are prevalent in society. The signatures of these images can be extracted as sensor pattern noise (SPN) and classified according to their source devices. In this paper, the authors assess the reliability of an unsupervised classifier for forensic investigation of digital images recovered from storage devices and to identify the best position for cropping the images before processing. Cross validation was performed on the classifier to assess the error rate and determine the effect of the size of the sample space and the classifier trainer on the performance of the classifier. Moreover, the authors find that the effect of saturation and subsequently the contamination of the SPN in the images affected performance negatively. To alleviate the negative performance, the authors identify the areas of images where less contamination occurs to perform cropping.