Detecting the Composite of Photographic Image and Computer Generated Image Combining with Color, Texture and Shape Feature

With the development of computer graphic techniques and smaller visual difference between photographic images (PG) and computer graphics (CG), splicing of computer graphics and photographic is becoming more common, which causes the need for automatically distinguishing computer generated images from real photographs. Based on several visual features that derived from color, texture and shape feature, using posterior probability support vector machine (PPSVM), this paper presents a method for classification of photographic image and computer generated images, and detection of local forgery composite of them. These features are obtained from 800 computer generated images and 800 photographs, and used to train and test the image samples with the PPSVM. The PPSVM can calculate the probability of image belonging to photographic images or computer generated images. We can achieve the result of classification by setting a threshold of probability. The classification accuracies for color, texture and shape features are 76.75%, 85.25% and 69% respectively. The accuracy is improved to 89.375% with combined color, texture and shape features. The proposed classification method is used to detect the local forgeries composite of the CG elements in photographic, and vice visa. Experimental results show that the proposed method is efficient with good detection rate of local forgeries composite of the PG and the CG. It also possess low dimensional features and low time complexity.