史二颖,朱家群,杨长春.基于最近邻搜索耦合近邻损耗聚类的图像伪造检测算法[J].包装工程,2018,39(5):185-190.
SHI Er-ying,ZHU Jia-qun,YANG Chang-chun.Image Forgery Detection Algorithm Based on Nearest Neighbor Search Coupling Neighbor Loss Clustering[J].Packaging Engineering,2018,39(5):185-190. |
| 基于最近邻搜索耦合近邻损耗聚类的图像伪造检测算法 |
| Image Forgery Detection Algorithm Based on Nearest Neighbor Search Coupling Neighbor Loss Clustering |
| 投稿时间：2017-04-22 修订日期：2018-03-10 |
| DOI：10.19554/j.cnki.1001-3563.2018.05.035 |
| 中文关键词: 图像伪造检测 最近邻搜索 SURF特征 KD树 特征聚类 Haar小波响应 |
| 英文关键词: image forgery detection nearest neighbor search SURF feature KD tree feature clustering Haar wavelet response |
| 基金项目:江苏省自然科学基金（BK20140159）；江苏省自然科学基金（BK20135638） |
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 目的 为了解决当前图像伪造检测算法在对图像进行伪造检测时，主要依靠全局搜索的方式来完成特征点匹配，导致其检测效率较低，且在对复杂伪造图像进行检测时，易出现检测精度不高和检测错误的不足。方法 提出基于最近邻搜索耦合近邻损耗聚类的图像伪造检测算法。首先引入积分图像的方法，对图像进行预处理，借助Hessian矩阵行列式来提取特征点。利用特征点构建圆形区域，通过求取圆形区域内Haar小波响应获取特征点的特征描述符。然后通过特征描述符建立KD树索引，利用最近邻搜索方法代替SURF中全局搜索的方法，对SURF进行改进，完成特征点的匹配。最后，利用特征点间的近邻关系求取近邻函数值，通过近邻函数值对特征点进行聚类，完成图像的伪造检测。结果 实验结果显示，与当前图像伪造检测算法相比，所提算法具有更高的检测效率以及更高的检测正确度。结论 所提算法具备较高的检测精度，在印刷防伪与信息安全等领域具有较好的应用价值。 |
| 英文摘要: |
| The work aims to solve the problem that the current image forgery detection algorithm matches the feature points by mainly relying on the global search during the image forgery detection, thus leading to low detection efficiency; and to solve such deficiencies as low detection accuracy and detection error likely to occur during the detection of complex forgery image. An image forgery detection algorithm based on the nearest neighbor search coupling neighbor loss clustering was proposed. Firstly, the image was pre-processed by the method of integral image, and then the feature points were extracted by Hessian matrix determinant. The feature points were used to construct the circular region, and the feature descriptors of the feature points were obtained by calculating the Haar wavelet response in the circular region. Then, the KD tree index was established by the feature descriptor, and the nearest neighbor search method was used instead of the global search method in SURF to improve the SURF and complete the matching of feature points. Finally, the nearest neighbor function value was obtained with the nearest neighbor relation between the feature points, and then the feature points were clustered by the nearest neighbor function values, and the image forgery detection was completed. The experimental results showed that the proposed algorithm had higher detection efficiency and higher detection accuracy compared with the current image forgery detection algorithm. The proposed algorithm has high detection accuracy and good application value in the field of printing anti-counterfeiting and information security. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
渝公网安备 50010702501716号