首页> 外文会议>Iberian conference on pattern recognition and image analysis >Automatic Tampering Detection in Spliced Images with Different Compression Levels
【24h】

Automatic Tampering Detection in Spliced Images with Different Compression Levels

机译:具有不同压缩级别的拼接图像中的自动篡改检测

获取原文

摘要

In this paper, we introduce a blind tampering detection method based on JPEG ghosts [3] capable of detecting tampering when it is created by splicing regions with different compression levels in an image. Given an image, a set of re-compressions of that image is generated and used to extract a feature vector to train a Support Vector Machine classifier. We used two different datasets in our experiments. The first one, extracted from Columbia Uncompressed Image Splicing Detection, was used to compare results with other works. Our method outperformed the previous ones when dealing with small tampered regions and similar qualities, offering hit rates above 97% (100% in the case of non-tampered images). With the second dataset, CASIA1 Tampered Image Detection Evaluation Dataset, our method offered a hit rate of 98.71% when discerning between the original and the spliced image, with just a 0.44% of non-tampered images wrongly classified.
机译:在本文中,我们介绍了一种基于JPEG虚影的盲目篡改检测方法[3],该方法可以通过在图像中拼接具有不同压缩级别的区域来创建篡改时检测篡改。给定一个图像,将生成该图像的一组重新压缩,并用于提取特征向量以训练支持向量机分类器。我们在实验中使用了两个不同的数据集。第一个是从哥伦比亚未压缩图像拼接检测中提取的,用于将结果与其他作品进行比较。当处理较小的篡改区域和类似质量时,我们的方法优于以前的方法,提供了97%以上的命中率(对于未篡改的图像,则为100%)。使用第二个数据集CASIA1篡改图像检测评估数据集,当我们区分原始图像和拼接图像时,我们的方法提供了98.71%的命中率,而只有0.44%的未篡改图像被错误分类。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号