基于SVD和直方图的JPEG图像篡改盲检测算法

现有的大多数图像篡改检测算法不能够很好地检测多次篡改区域,针对此不足,提出了一种有效的基于SVD和直方图的JPEG图像篡改盲检测算法。该算法首先以设定的窗口块在待检测图像上依次滑动一个像素得到每个单独的滑窗分块,每个分块用奇异值分解(SVD)值表征;然后字典排序所有分块量化后的SVD值矩阵,并通过统计排序后的矩阵的偏移频率来得到直方图;最后通过直方图设定阈值以判断分块是否属于复制粘贴块。实验结果表明,该算法不仅能对单次篡改区域进行准确定位,还能较好地检测到多次篡改区域。

Jpeg image tampering blind detection algorithm

As to the most image forgery detection algorithms, it is hard to detect multi-tampered regions, thus an effective algorithm for JPEG image forgery blind detection based on SVD and histogram is proposed. First, moving the pixel with a specified size window in the input image one by one and obtain every single patch, which will be represented by SVD value respectively. Then quantify all of the patches to get the matrix sorted by dictionary and calculate the histogram using the deviation frequency from that matrix. Finally, verifying the copy-move regions combining histogram and threshold value. Experimental results show that for single-tampered regions this algorithm can not only stand out in terms of precision in patch location, but also performance well for multi-tampered regions.