Passive image splicing detection based on third order statistical features
基于三阶统计特征的被动图像拼接检测

This paper proposed a third order statistical features based method to detect image splicing operation passively. The dependences among neighboring three states in the state-array were modeled as CCPM which was treated as discriminative features for SVM classification. Since the dimensionality of higher order statistical features grows exponentially with the order, it introduced PCA to decrease the complexity for classification and to overcome the potential over-fitting problem. Experimental results show that the conditional co-occurrence probability matrix features outperform traditional Markov features and gray level co-occurrence matrix features in both 8×8 block DCT domain and spatial domain. PCA is an effective tool for image splicing detection and new features with much fewer dimensionalities after PCA perform as good as original features.