Study Of Image Splicing Detection Based On Entropy Featureand Markov Feature

Digital images have become an important way for people to send andacquire information because of its intuitive and more convincingadvantages. With the popularity of digital products, digital images are seeneverywhere, and image editing software makes digital images more easilymodified. Image editing software brings us convenience, as well assecurity risks. So the authenticity detection of digital images has attract aincreasing attention recently.Image splicing, which is a process of copying a part of a image andthen pasting into the same or another image, is one of the typical way inimage tampering schemes. Some effective splicing approaches have beenproposed recently, and splicing detection based on Markov feature andclassifier, which has comparatively higher detection accuracy, is a typicalway. However, Markov feature is the one-step transition probability matrixof difference array of image DCT coefficients, and it only analyses thecorrelation between the two adjacent elements in difference array, so thereis room for improvements. In this paper, we propose an improvedmulti-step Markov feature based on the typical Markov feature. Multi-stepMarkov feature is the multi-step transition probability matrix of differencearray of image DCT coefficients, and it contains the correlation betweenthe continuous elements in difference array. Experimental results show thatMulti-step Markov feature outperforms Markov feature, and that averagedetection accuracy based on multi-step Markov feature achieves89.12%.In order to further improve the detection accuracy, entropy feature isproposed in image splicing detection. In this paper, we proposed a way ofimage splicing detection based on entropy and multi-step Markov feature. Multi-step Markov feature extracts splicing traces in block discrete cosinetransform domain, and entropy feature extracts splicing traces in discretewavelet transform domain. The combination of entropy feature andmulti-step Markov feature can analyses the splicing traces from a differentangle, and increases the average detection accuracy up to90.20%. In orderto deduce the dimensionality of the combined features, genetic algorithm isutilized to optimize the combined features. Experimental results show thatgenetic algorithm can both deduce the dimensionality of the combinedfeatures and keep comparative detection accuracy with the originalfeatures.