基于目标检测约束的图像过滤方法研究  

Research on Image Filtering Method Based on Object Detection Constraints

作　　者：刘亚奇 张一凡 LIU Yaqi;ZHANG Yifan(Beijing Electronic Science and Technology Institute,Beijing 100070,P.R.China)

机构地区：[1]北京电子科技学院,北京市100070

出　　处：《北京电子科技学院学报》2024年第3期95-103,共9页Journal of Beijing Electronic Science And Technology Institute

摘　　要：图像过滤方法通过分析和匹配图像特征,检索得到可疑篡改来源图像,分析篡改图像和篡改来源图像的派生关系。本文提出一种基于目标检测约束的图像过滤方法,首次整合了目标检测算法和分布式加速的鲁棒特征索引,一方面利用目标检测算法使得特征检索范围得到约束,另一方面采用分布式局部特征提取方法,有效提高了图像特征检索的效率和准确度。此外,使用尺度可变的有约束图像拼接检测与定位网络在图像特征检索基础上进行深度匹配,进一步对篡改区域和篡改来源区域进行精确查找。实验结果证明,本文算法在篡改图像检索准确性、查询效率方面有明显改进。In image filtering methods,image features are analyzed and matched to identify the potential tampering sources and then to analyze the derivative relationship between the tampered images and their sources.In this paper,we propose a novel image filtering method based on object detection constraints by integrating the object detection(YOLO v5)and the distributed SURF indexing,where the object detection is adopted to constrain the feature retrieval scope and the distributed local feature extraction is utilized to improve the efficiency and accuracy of image feature retrieval.In addition,a scalable con⁃strained image splicing detection and localization network is employed for deep matching on the basis of image feature retrieval,allowing for precise identification of tampered areas and their sources.Experiment results demonstrate that the proposed method achieves significant improvements in tampered image retrieval accuracy and query efficiency.

关 键 词：图像取证 数字图像篡改定位 图像来源分析 图像拼接篡改区域定位 深度学习 

分 类 号：TN919[电子电信—通信与信息系统]