利用CSURF的图像复制粘贴篡改检测

Image Copy-Move Forgeries Detection Using CSURF

现有的复制粘贴篡改检测算法大多是针对灰度图像设计的,并没有利用图像的颜色信息.在篡改区域纹理信息较少时,传统的基于关键点的算法无法提取关键点以识别篡改区域.为此,提出了一种颜色不变性特征描述子CSURF(colored speeded-up robust feature)并应用于篡改检测.提出的算法首先提取图像的颜色梯度,利用颜色梯度代替灰度图像提取关键点,通过计算描述子点积反余弦提高关键点匹配的运算效率,并对匹配结果进行几何变换估计以呈现篡改区域.提出的算法在篡改区域纹理信息较少时,也能利用颜色信息提取关键点,以识别篡改区域,并且对多种后处理操作及退化具有较好的鲁棒性.

Most of the existing copy-move forgery detection methods are designed for grayscale images without any color information used. However,when the texture information of the duplicated regions is very little,the traditional keypoint-based method cannot extract keypoints to identify the duplicated regions. To solve this problem,a novel color invariant feature descriptor colored speeded-up robust feature（CSURF）is proposed for detecting image copy-move forgeries. The proposed method first extracts image color gradient and usesthe gradient to extract keypoints instead of the grayscale image. Then,inverse cosine of the dot product between descriptors is calculated to improve efficiency. In the end,geometrical estimation is performed to expose the duplicated regions. The proposed method can use color information to extract keypoints when the texture information of the tampered regions is little. Moreover,it has better robustness against various post-processing operations and degradations.