改进的卷积神经网络用于对比度增强取证

摘要

提出一种改进的卷积神经网络(modified convolutional neural network,MCNN)用于图像的对比度增强取证.其中MCNN第1层是预处理层,这一层将输入图像转化为二值灰度共生矩阵(binary gray-level co-occurrence matrix,BGLCM),其余各层与传统的卷积神经网络相同,这些层可从BGLCM上学习特征并以此进行分类.该方法的特征提取和分类可同时进行优化,使提取到的特征更适合对比度增强检测.实验表明,所提方法不仅可检测传统的对比度增强技术和两种反取证技术产生的对比度增强图像,还可区分对比度增强时所采用的参数.%Contrast enhancement forensics has drawn much attention in image forensics recently.This paper proposes a modified convolutional neural network (MCNN).The first layer in the MCNN is a preprocessing layer,which converts an input image to a binary gray-level co-occurrence matrix (BGLCM).The other layers are the same with conventional CNN,which learn features from BGLCM for classification.Compared with previous methods,feature extraction and classification can be optimized simultaneously,making the extracted features more suitable for contrast enhancement detection.Experimental results show that the proposed method can detect contrast enhanced images produced by conventional contrast enhancement and two anti-forensic techniques.It can also distinguish parameters used in the contrast enhancement.