基于DeepLab v3+的多任务图像拼接篡改检测算法

在图像拼接篡改检测任务中，受篡改区域尺度多样性及模糊操作的影响，传统分类算法难以提取图像篡改特征。提出一种基于DeepLab v3+的图像拼接篡改检测算法，使用浅层图像特征预测图像的篡改区域边界，提高模型对篡改边界的敏感性。在此基础上，通过多尺度融合特征对图像篡改区域进行分割，并在原空洞空间金字塔模块中融合空间和通道注意力机制，从而提高模型对多尺度篡改区域的适应性。实验结果表明，所提算法能有效检测图像的篡改区域，在CASIA v1.0和Columbia数据集中的分割精度分别为0.754 6和0.727 8，优于DCT、BAPPY、MFCN等算法。

Multi-task Algorithm for Image Splicing Forgery Detection Based on DeepLab v3

In the detection of image splicing forgery, it is difficult for the traditional classification algorithms to extract the tampering features of the image due to the scale diversity of the tampered area and the interference of the fuzzy operation.In order to solve this problem, a multi-task algorithm based on Deeplab v3+ is proposed for detecting image splicing forgery.The algorithm uses the shallow image features to predict the boundary of the tampered area, so the sensitivity of the model to the tampered area boundary is improved.On this basis, multi-scale fused features are used to segment the tampered area in the image.The spatial and channel attention mechanisms are integrated in the dilated spatial pyramid module to improve the adaptability of the model to multi-scale tampered areas.The experimental results show that the improved algorithm displays a segmentation accuracy of 0.754 6 on the CASIA v1.0 dataset and 0.727 8 on the Columbia dataset, outperforming DCT, BAPPY, MFCN and other advanced algorithms.