Beijing Key Lab of Intelligent Telecommunication Software and Multimedia, Beijing University of Posts and Telecommunications, Beijing 100876, China.
School of Computer Science and Software Engineering, University of Science and Technology Liaoning, Anshan 114051, China.
Comput Intell Neurosci. 2019 Nov 11;2019:9492358. doi: 10.1155/2019/9492358. eCollection 2019.
Seam carving is an excellent content-aware image resizing technology widely used, and it is also a means of image tampering. Once an image is seam carved, the distribution of magnitude levels for the pixel intensity differences in the local neighborhood will be changed, which can be considered as a clue for detection of seam carving for forensic purposes. In order to accurately describe the distribution of magnitude levels for the pixel intensity differences in the local neighborhood, local neighborhood magnitude occurrence pattern (LNMOP) is proposed in this paper. The LNMOP pattern describes the distribution of intensity difference by counting up the number of magnitude level occurrences in the local neighborhood. Based on this, a forensic approach for image seam carving is proposed in this paper. Firstly, the histogram features of LNMOP and HOG (histogram of oriented gradient) are extracted from the images for seam carving forgery detection. Then, the final features for the classifier are selected from the extracted LNMOP features. The LNMOP feature selection method based on HOG feature hierarchical matching is proposed, which determines the LNMOP features to be selected by the HOG feature level. Finally, support vector machine (SVM) is utilized as a classifier to train and test by the above selected features to distinguish tampered images from normal images. In order to create training sets and test sets, images are extracted from the UCID image database. The experimental results of a large number of test images show that the proposed approach can achieve an overall better performance than the state-of-the-art approaches.
seam carving 是一种广泛使用的优秀的内容感知图像缩放技术，也是一种图像篡改手段。一旦对图像进行 seam carving，像素强度差异的局部邻域的幅度级别的分布就会发生变化，这可以被认为是用于取证目的的 seam carving 检测的线索。为了准确描述局部邻域的像素强度差异的幅度级别的分布，本文提出了局部邻域幅度出现模式（LNMOP）。LNMOP 模式通过统计局部邻域中幅度级别的出现次数来描述强度差的分布。在此基础上，本文提出了一种用于图像 seam carving 的取证方法。首先，从 seam carving 伪造检测的图像中提取 LNMOP 和 HOG（方向梯度直方图）的直方图特征。然后，从提取的 LNMOP 特征中选择最终用于分类器的特征。提出了基于 HOG 特征层次匹配的 LNMOP 特征选择方法，通过 HOG 特征级别确定要选择的 LNMOP 特征。最后，利用支持向量机（SVM）作为分类器，通过上述选择的特征进行训练和测试，以区分篡改图像和正常图像。为了创建训练集和测试集，从 UCID 图像数据库中提取图像。大量测试图像的实验结果表明，所提出的方法的整体性能优于最新方法。
