基于改进YOLOX-X的图像篡改检测算法

Image Tampering Detection Algorithm Based on Improved YOLOX-X

在图像篡改检测任务中,重点需要关注的是伪造区域与原始区域的差别特征,目前的图像篡改检测算法普遍存在检测效果较差、识别精度不题。为解决上述问题,文章提出一种基于改进YOLOX-X的图像篡改检测算法。为了将提取到的多尺度篡改特征进行更充分的融合,引入了ASFF特征融合机制,并且对本篡改检测算法中的损失函数进行了改进,使用EIOU对损失函数进行优化,使网络更快速准确的识别定位篡改区域。通开的图像篡改数据集上的实验结果表明,本文提出方法的整体性能要优于其他主流的图像篡改检测算法。

In the task of image tamper detection,the focus is on the difference between the forged area and the original area.rent image tamper detection algorithms generally have the problems of poor detection effect and low recognition accuracy.To ve these problems,this paper presents an improved image tampering detection algorithm based on YOLOX-X.In order to fuse the racted multiscale tamper features more fully,an ASFF adaptive feature fusion mechanism is introduced,and the loss function this tamper detection algorithm is improved.The loss function is optimized by using EIOU to make the network identify and ate tamper areas more quickly and accurately.The experimental results on an open image tamper dataset show that the overall formance of the proposed method is better than other mainstream image tamper detection algorithms.