摘要
为了解决当前篡改图像的检测算法主要依靠单一的特征进行描述以及欧几里德距离进行匹配,导致篡改图像的检测率较低的问题,以及在对图像复制粘贴后进行一系列后处理操作的篡改图像检测时,容易出现匹配错误和鲁棒性差的问题,采用一种多距离特征匹配的篡改图像检测算法。首先,对获取到的图像提取尺度不变特征变换(SIFT)特征,在SIFT特征待描述区域的基础上,提取具有权值旋转不变均匀性的局部二值模式(LBP)特征,构成特征描述子;其次,分别计算描述子之间的标准欧几里德距离、相关距离以及汉明距离,通过多距离匹配改进g2nn算法进行特征的初次匹配;最后,通过凝聚型分层特征聚类以及随机一致性(RANSAC)算法去除存在的错误匹配点,完成篡改图像的检测。在MICC-F220图像数据库上进行了测试,实验结果表明,与当前2种主流算法相比,总体准确率分别提高了2.86%和2.11%,对于缩放、旋转以及缩放+旋转的后处理均具有很好的鲁棒性,是一种研究复制粘贴后进行缩放和旋转后处理的篡改图像检测的有效方法。
In order to solve the problem that the detection algorithm of the current tampering image relies mainly on a single feature to describe and the Euclidean distance to match, the detection rate of the tampering image is comparatively low, and matching errors and poor robustness are prone to occur when a series of post processing tampering images are detected after copy paste images, a multi-distance feature matching detection algorithm is used in this paper. Firstly, the scale-invariant feature transform(SIFT) feature was extracted from the acquired image, and the local binary patterns(LBP) feature with the rotation invariance uniformity of the weight was extracted on the basis of the SIFT region to be described, so the feature descriptor was constructed. Secondly, the standard Euclidean distance, correlation distance and hamming distance were calculated respectively, and G2 NN algorithm was improved by multi-distance matching to perform the initial match of the feature. Finally, mismatched points were removed by condensed hierarchical feature clustering and random sample consensus(RANSAC) algorithm to complete tampering image detection.The test was carried out on the MICC-F220 image database with the result that the overall accuracy of the proposed algorithm is improved by 2.86% and 2.11% respectively compared with the two mainstream algorithms available. It is robust to scaling, rotation and scaling and rotation post-processing. It is an effective method to detect tampering image detection after copying and pasting and then performing scaling and rotation processing.
作者
张威虎
郑佳雯
郭明香
陶智慧
贺元恺
ZHANG Wei-hu;ZHENG Jia-wen;GUO Ming-xiang;TAO Zhi-hui;HE Yuan-kai(College of Communication and Information Engineering,Xi'an University of Science and Technology,Xi'an 710054,China)
出处
《西安科技大学学报》
CAS
北大核心
2019年第4期665-671,共7页
Journal of Xi’an University of Science and Technology
基金
陕西省自然科学基金(2017JM6102)
