基于内容感知图像缩放的取证技术研究

Study on Forensic Techniques Based on Content-aware Image Reszing

近几年,内容感知的智能图像缩放技术得到了普遍重视和广泛应用,其中最成熟和流行的细缝裁剪(SeamCarving)技术可以在尽量保持图像细节不扭曲的情况下进行图像缩放,该技术还可以方便地应用于图像细节的篡改。提出了一种基于马尔科夫特征的细缝裁剪篡改数字图像识别算法,将图像进行8×8分块,对每块进行DCT变换,对所得DCT系数矩阵求取相邻系数水平和竖直方向的差矩阵,利用2个方向的差矩阵得到基于马尔科夫转移概率矩阵的图像特征,并将得到的特征利用支持向量机进行分类训练并对正常图像和细缝裁剪篡改图像进行识别分类。实验结果表明,算法性能优异,可以有效识别正常图像和细缝裁剪篡改图像。

The content-aware image resizing techniques are gaining extensive attention and being widely used these recent years.The Seam-Carving is the most mature one among these techniques,it can resize the image without distorting the materials in image,and also be easily used to tamper the materials in image. To deal with the Seam-Carving tampered image,a detection algorithm based on Markov features is proposed in this paper. The algorithm divides the image into 8 × 8 non-overlapping blocks,performs DCT on every block,gets difference matrix in horizontal and vertical direction from adjacent coefficients in DCT blocks,and then finally obtains features based on Markov transition probability matrix. These features are trained by SVM and then used to identify the seam-carved images from normal images. The experiment results show that the algorithm is outstanding and can be used to identify Seam-Carved image from normal image effectively.