An image splicing forgery detection algorithm based on a dual-label supervised dual-stream ringed residual u-net(RRU-Net) was proposed to address existing issues such as blurred edges and neglected background information in current methods. Firstly, building upon the RRU-Net, a mask edge stream was designed. Bidirectional interaction between the mask stream and the mask edge stream was achieved through an adaptive layer, which fully utilized their complementary information to enhance the clarity of the detected mask edges. Secondly, at the input stage, the input image was converted from the RGB color space to the YCrCb color space, allowing the luminance and chrominance information to be analyzed separately to reveal potential tampering traces. Subsequently, a coordinate attention mechanism was incorporated into the intermediate layers of the network to recalibrate the fused features, thereby enhancing the focus on critical features. Finally, within the mask stream, a combination of binary cross entropy loss(BCE) and Dice loss functions was employed to mitigate the negative impact of foreground-background (area) imbalance in the samples. The experimental results demonstrate that compared to the baseline model RRU-Net, the F1-score of the proposed method is increased by 25.96%, 5.15%, 7.14%, and 9.96% on the public datasets CASIA1.0, CASIA2.0, IMD2020, and the self-constructed synthetic dataset, respectively. Moreover, the comprehensive detection performance of the proposed method is found to be significantly superior to all other compared methods in the study. Furthermore, in robustness experiments, the proposed method is subjected to post-processing operations such as JPEG compression and image scaling. Although a slight decrease in F1-score is observed, the method generally outperforms the comparative approaches, demonstrating strong robustness. The superior detection performance of the method is further validated.
评估指标:图像拼接篡改检测,本质上是一个像素级的二值分类问题,因此采用F1和交并比(intersection over union,IoU)对模型进行评估,F1是一个评估二分类模型性能的指标,由查准率(Precision)和查全率(Recall)综合计算得出。IoU是一个用于量化两个区域重叠程度的度量。F1和IoU的计算公式分别为
将RRU-Net作为基础模型,其参数量和计算复杂度适中。引入双流结构后,虽然参数量和浮点运算次数(floating point operations per second,FLOPs)有所增加,但模型性能得到大幅提升。YCrCb转换层为纯计算单元,不包含训练参数,故不占用模型存储空间。CA注意力模块作为一个轻量级组件,参数量仅0.051 1 MB,FLOPs低至0.001 6 G,可无缝集成到主体网络中,且几乎不增加额外开销,各模块具体参数量和FLOPs如表2所示。
JinShun, GeDongyuan, YaoXifan. Image stitching based on matching point increment and random sample consensus[J]. Science Technology and Engineering, 2025, 25(6): 2435-2441.
YangYanyu, WeiWeimin, ZhangYunqin. An improved dual-stream Faster R-CNN model for image tampering detection[J]. Computer Applications and Software, 2023, 40(12): 189-194.
[20]
LiangE, ZhangK, HuaZ, et al. BASNet: boundary assisted network for image splicing forgery detection[J]. IEEE Transactions on Multimedia, 2025, 27: 8002-8015.
[21]
PengZ, HuangW, GuS, et al. Conformer: Local features coupling global representations for visual recognition[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. New York: IEEE, 2021: 367-376.
[22]
HouQ, ZhouD, FengJ. Coordinate attention for efficient mobile network design[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville: IEEE, 2021: 13713-13722.
[23]
HuJ, ShenL, SunG. Squeeze-and-excitation networks[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 7132-7141.
[24]
WooS, ParkJ, LeeJ Y, et al. CBAM: convolutional block attention module[C]// Proceedings of the European conference on computer vision (ECCV). Berlin: Springer, 2018: 3-19.
[25]
DongJ, WangW, TanT. Casia image tampering detection evaluation database[C]// IEEE China Summit and International Conference on Signal and Information Processing. Beijing: IEEE, 2013: 422-426.
[26]
NovozamskyA, MahdianB, SaicS. IMD2020: a large-scale annotated dataset tailored for detecting manipulated images[C]// Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision Workshops. Snowmass: IEEE: 2020: 71-80.
[27]
BappyJ H, SimonsC, NatarajL, et al. Hybrid LSTM and encoder-decoder architecture for detection of image forgeries[J]. IEEE Transactions on Image Processing, 2019, 28(7): 3286-3300.
[28]
LinT Y, MaireM, BelongieS, et al. Microsoft coco: common objects in context[C]// Computer Vision-ECCV 2014: 13th European Conference. Zurich: Springer International Publishing, 2014: 740-755.
PeiR, WeiW, YuX, et al. Image tampering localization based on dual-stream feature fusion[C]// International Conference of Pioneering Computer Scientists, Engineers and Educators. Singapore: Springer, 2024: 295-305.
WuJinghui, YanCaiping, LiHong, et al. Edge-guided dual attention image stitching detection network[J]. Journal of Image and Graphics of China, 2024, 29(2): 430-443.
[34]
ZhangY, ZhuG, WuL, et al. Multi-task SE-network for image splicing localization[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2021, 32(7): 4828-4840.
[35]
BiX, WeiY, XiaoB, et al. RRU-Net: The ringed residual U-Net for image splicing forgery detection[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Long Beach: IEEE, 2019: 30-39.
[36]
WuY, AbdAlmageedW, NatarajanP. Mantra-Net: manipulation tracing network for detection and localization of image forgeries with anomalous features[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019: 9543-9552.
LiuLiying, WangJinxin, CaoShaoli, et al. U-Net for detecting small manipulated regions[J]. Journal of Image and Graphics, 2022, 27(1): 176-187.
[39]
ZhuangP, LiH, TanS, et al. Image tampering localization using a dense fully convolutional network[J]. IEEE Transactions on Information Forensics and Security, 2021, 16: 2986-2999.
XuYue, YuanChengsheng, LiuQingcheng, et al. Multi-scale perceptual learning for image manipulation detection and localization[J]. Computer Systems and Applications, 2025, 34(7): 117-127.
[42]
AkramA, JaffarM A, RashidJ, et al. CMV2U-Net: AU-shaped network with edge-weighted features for detecting and localizing image splicing[J]. Journal of Forensic Sciences, 2025, 70(3): 1026-1043.