基于均值漂移的图像复制粘贴伪造盲检测

摘要：随着数字多媒体技术及计算机网络技术的发展，数字图像在信息技术时代扮演着越来越重要的角色，图像的真实性成为现代人们广泛关注的热点之一，为此提出了一种基于均值漂移的图像复制粘贴伪造盲检测算法。提取图像的SURF（Speed up robust feature）特征点，通过最近邻匹配方法进行特征匹配，滤除冗余点，初步定位复制粘贴伪造区域。均值漂移（Mean Shift）将具有相同或相似属性的图像像素分割为同一区域，利用匹配后的SURF特征点与其所在均值漂移分割区域的位置依赖关系确定伪造区域，并采用边缘直方图和HSV颜色直方图衡量特征点所在分割区域与相邻分割区域间的相似度，进一步细化伪造检测结果，最终实现图像的复制粘贴伪造盲检测。实验结果表明，该算法能够鲁棒地、高效地检测出图像的复制粘贴伪造区域。

Copy-paste image forgery blind detection based on mean shift

The traditional blind detection methods of image copy-paste forgery are time consuming, of high computation cost and low detection precision. A blind detection algorithm of copy-paste image forgery based on Mean Shift (MS) was proposed in this paper, which extracted Speed Up Robust Feature (SURF) points and then performed feature matching utilizing the method of best bin first in order to filter redundant points and locate the copy-paste forgery regions preliminarily. Pixels with the same or similar attributes would be segmented in the same region after implementing MS. The copy-paste regions could be detected according to the position dependency between matched feature point with its segmented region of MS and the detection result would be further refined by comparing the similarity of edge histogram and HSV (Hue-Saturation-Value) color histogram among the segmented regions of matched SURF and its neighborhood, and those regions with large similarity were included in the forged region. The experimental results show that the copy-paste forgery regions are detected accurately in the image with clear outline and rich details, and the proposed algorithm can robustly and efficiently detect the copy-paste forgery regions of image.