基于深度特征提取和图神经网络匹配的图像复制粘贴篡改检测

针对图像中特征提取不均匀、单尺度超像素划分对伪造定位结果影响较大的问题,提出一种基于深度特征提取和图神经网络(graph neural network,GNN) 匹配的图像复制粘贴篡改检测(cope-move forgery detection,CMFD) 算法。首先将图像进行多尺度超像素分割并提取深度特征,为保证特征点数目充足,以超像素为单位计算特征点分布的均匀度,自适应降低特征提取阈值；随后引入新的基于注意力机制的GNN特征匹配器,进行超像素间的迭代匹配,且用随机采样一致性(random sample consensus,RANSAC) 算法消除误匹配；最后将多尺度匹配结果进行融合,精确定位篡改区域。实验表明,所提算法具有良好的性能,也证明了GNN在图像篡改检测领域的可用性。

Image copy-move forgery detection based on depth feature extraction and graph neural network matching

An image copy-move forgery detection (CMFD) algorithm based on depth feature extraction and graph neural network (GNN) matching is presented to solve the problem that feature extraction is not uniform in image and single-scale superpixel segmentation has a strong impact on the result of forged location.Firstly,the image is segmented into multi-scale superpixels and the depth features are extracted.To ensure sufficient number of feature points,the uniformity of feature point distribution is calculated in the unit of superpixels,and the feature extraction threshold is reduced adaptively.Then a novel GNN feature matcher based on attention mechanism is introduced to perform iterative matching between superpixels,and random sample consensus (RANSAC) algorithm is used to eliminate mismatching.Finally,the multiscale matching results are fused to accurately locate the tampered areas.Experiments show that the proposed algorithm has good performance and the applicability of the GNN in the field of image tampering detection.