基于梯度增强的面部篡改检测算法

Study on Facial Tamper Detection Algorithm Based on Gradient Enhancement

面部篡改检测在舆情防控、司法取证等方面起着十分重要的作用。近年来,卷积神经网络在面部篡改检测领域表现出了强大的能力。然而,传统的卷积操作将感知区域的图像灰度值信息直接聚合计算,忽略了图像的细粒度梯度信息,使得检测效果难以进一步提升。基于以上考虑,本文提出一种基于梯度增强的面部篡改检测算法。算法实现了一种基于中心差分卷积的梯度增强模块,通过逐模块的特征增强以弥补传统卷积操作的不足。在FaceForensics++数据集上进行实验验证,显示算法具有更强的细粒度表征能力,同时实验结果较主流算法取得了最高2%的平均准确度提升。

Facial tampering detection plays a vital role in public opinion supervision,judicial evidence collection.In recent years,convolution neural network has shown significant advantages in the field of facial tamper detection.However,the traditional convolution operation directly aggregates the gray value information of the image in the sensing area,neglecting the fine-grained gradient information of the image,which renders inaccurate detection results.In view of the above-mentioned issues,a face tamper detection algorithm based on gradient enhancement is proposed in this paper,which implements a gradient enhancement module based on central difference convolution,and can make up for the shortcomings of traditional convolution operation by module by module feature enhancement.Experimental validation on FaceForensics++dataset show that the proposed algorithm has superior fine-grained characterization capability,and the experimental results have achieved an average accuracy improvement of up to 2%compared with the mainstream algorithms.