基于特征融合与局部异常的图像篡改检测  

Image Tampering Detection Based on Feature Fusion and Local Anomalies

作　　者：张虹 陈赵乐 董方敏[1,2] 孙水发[1,2] ZHANG Hong;CHEN Zhao-le;DONG Fang-min;SUN Shui-fa(College of Computer and Information Technology,China Three Gorges University,Hubei Yichang 443002,China;Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering,China Three Gorges University,Hubei Yichang 443002,China)

机构地区：[1]三峡大学计算机与信息学院,湖北宜昌443002 [2]三峡大学湖北省水电工程智能视觉监测重点实验室,湖北宜昌443002

出　　处：《计算机仿真》2024年第11期269-277,共9页Computer Simulation

基　　金：国家自然科学基金资助项目(61871258)资助。

摘　　要：数字图像处理技术的不断发展使得图像篡改检测任务越来越具有挑战性。图像篡改检测方法往往需对整个图像进行无差异检测,极大程度地增加了模型学习的复杂度;此外不同的图像大小和采样、卷积、池化等操作也会丢失部分图像的空间信息和篡改特征。为尽量减少特征信息丢失,文中所提的模型通过对上采样和下采样的对应特征进行融合以及将噪声流和约束卷积流与模型的卷积特征进行融合,保存图像的低级语义信息;其次,基于分割模型构建了一个局部异常区域检测机制,重点关注图像异常区域,减少模型复杂度;最后利用注意力增强对篡改特征的关注,解决语义信息不明确的问题。提出的篡改检测方法在三个基准数据集上取得了最优性能,且与最新方法进行了对比试验,在CASIA数据集上F1值提升了15.9%,NIST2016数据集提升了4.5%,COVERAGE数据集提升了5.7%。The continuous development of digital image processing techniques has made the task of image tampe-ring detection increasingly challenging.In addition,different image sizes and operations such as sampling,convolution and pooling can also result in the loss of some spatial information and tampered features.In order to minimise the loss of feature information,the model in this paper preserves the low-level semantic information of the image by fusing the corresponding features of up sampling and down sampling and fusing the noise stream and constrained convolution stream with the convolution features of the model;secondly,a local abnormal region detection mechanism is construc-ted based on the segmentation model to focus on the abnormal regions of the image and reduce the complexity of the model;finally,attention is used to enhance the focus on the tampered features to solve the problem of unclear semantic information.The tampering detection method in this paper achieves optimal performance on three benchmark datasets and is tested against the latest methods,with F1 values improving by 15.9%on the CASIA dataset,4.5%on the NIST2016 dataset and 5.7%on the COVERAGE dataset.

关 键 词：图像篡改检测 特征融合 局部异常检测 注意力增强 约束卷积 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]