Classifying Computer Generated Graphics and Natural Images Based on Image Contour Information

摘要

Contemporary computer graphics rendering software is capable of generating highly photorealistic images, resulting in big challenges for image authentication. In this paper, we begin with a discussion on current issues of image authentication. We then introduce a new approach to distinguish Computer Graphics (CG) from photographic images based on image contour information. The novel method is motivated by a detailed analysis of the fundamental differences between the two image categories, derived from both their distinct generation processes and from natural image statistics. In addition, we investigate the effectiveness of diversified image color representations in computer graphics identification. We also evaluate the classification performance of various feature types in different sub-bands. The experiments show that the features of high frequency image components, extracted from Hue channel in HSV color space, demonstrate best overall performance.