Passive Detection of Image Splicing using Conditional Co-occurrence Probability Matrix

In the past few years second order statistical features (e.g. Markov transition probability matrix) have been proved to be effective features for image forgery detection. In this paper, conditional co-occurrence probability matrix (CCPM) which is third order statistical features is proposed to detect image splicing. Since the dimensionality depends exponentially on the order of features, principal component analysis is employed to overcome the high dimensionality introduced computational complexity and over-fitting for a kernel based supervised classifier. Experimental results show that conditional co-occurrence probability matrix demonstrates its effectiveness in block discrete cosine transform domain, it outperforms Markov features in both block discrete cosine transform domain and spatial domain, principal component analysis is proved to be an effective dimensionality reduction method for image splicing detection.