基于多残差学习与注意力融合的中值滤波检测

Median filtering detection based on multiple-residual learning with attention fusion

针对小尺寸JPEG压缩图像携带有效信息较少、中值滤波痕迹不明显的问题,提出一种基于多残差学习与注意力融合的图像中值滤波检测算法。该算法将多个高通滤波器与注意力模块相结合,获取带权值的多残差特征图作为特征提取层的输入,特征提取层采用分组卷积形式,对输入的多残差特征图进行多尺度特征提取,融合不同尺度的特征信息,同时采用密集连接方式,每一层卷积的输入来自前面所有卷积层的输出和。实验结果表明,针对小尺寸JPEG压缩图像的中值滤波检测,本文算法比现有算法具有更高的检测精度,且能更有效地检测与定位局部篡改区域。

A median filter detection algorithm based on multi-residual learning and attention fusion is proposed for small size JPEG compressed images carrying less effective information and less obvious median filter traces.In this algorithm, multiple high-pass filters are combined with the attention module to obtain multi-residual feature maps with weights as the input of the feature extraction layer, and the feature extraction layer adopts the form of grouped convolution to perform multi-scale feature extraction on the input multi-residual feature maps and fuse the feature information with different scales, while the dense connection is used, and the input of each convolution layer comes from the output sum of all previous convolution layers.The experimental results show that the proposed algorithm in this paper has higher detection accuracy than existing algorithms for median filter detection of small JPEG compressed images, and can detect and locate local tampered regions more effectively.