Detecting Doctored JPEG Image Based on Block Noise Analysis and Double JPEG Analysis

Abstract

Today we can take many pictures with smart phones or digital cameras, and can edit them easily by ourselves. These pictures are very useful not only for hobby, but for investigation. It is very important to check whether they are doctored or not. This paper proposes an automatic detection method of doctored JPEG images based on two different analyisis: the block noise analysis and the Double-JPEG analysis. Former can find unnatural boundaries of 8×8 DCT blocks while latter can find double saved images by other editing software. Finally SVM classifies images into doctored and undoctored groups based on the above analysis. Experimental results have shown that the detection accuracy of our method achieves 0.90 in terms of F-measure while J. He's method achieves 0.82.