원문정보
초록
영어
In this paper, we propose a camera-model detection method based on a hybrid approach. Varying camera inside imaging processing will lead to varying artifacts. A few of artifacts can reflect camera model-specific. To comprehensive track camera model-specific footprint, we build a hybrid approach by combining two-step Markov feature based model and CFA feature based model. A 132-D feature set is designed to perform camera-model classification. Images from seven camera models in the Dresden Image Database are chosen as our experiment database. Experiment results show that in seven models detection, the average detection accuracy of our method is 99.83%. Even the feature dimension is decreased to 40 by feature selection; its detection accuracy can still reach to 99.58%, which is higher than that of previous Markov method [12].
목차
1. Introduction
2. Hybrid Forensics Model for Detection
3. Markov Feature
3.1. Difference JPEG 2-D Array
3.2. Two-step Transition Probability Matrix with Threshold Setting
4. CFA Feature
5. Experiment
6. Conclusion
References