Distinguishing photorealistic computer graphics from natural images by imaging features and visual features

Although photorealistic computer graphics (PRCG) attempt to deceive our eyes, we can distinguish them from photographic images (PIM, i.e. natural images) depending on approaches of constructing discriminating features and using machine learning and data mining technique to obtain statistical regularity as the prior work in [3,4]. This paper is aiming to find the different contribution of different image components in identifying PIM and PRCG and present a new approach combining features from these separated components. Motivated by wavelet based image denoising, an image is divided into subimages corresponding to different image information by comparing all wavelet coefficients with a universal threshold. Whereafter imaging features and visual features respectively are extracted from different components. We compared the discrimination ability of these features and it is proved that the proposed approach using features from different components is effectual in distinguishing PIM and PRCG by our experimental results.