[1]孟 旭,孟 坤.基于神经网络的图像来源识别方法比较研究[J].计算机技术与发展,2022,32(01):111-116.[doi:10. 3969 / j. issn. 1673-629X. 2022. 01. 019]
MENG Xu,MENG Kun.Comparison and Analysis of Image Source IdentificationMethod Based on Neural Network[J].,2022,32(01):111-116.[doi:10. 3969 / j. issn. 1673-629X. 2022. 01. 019]
点击复制
基于神经网络的图像来源识别方法比较研究()
《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]
- 卷:
-
32
- 期数:
-
2022年01期
- 页码:
-
111-116
- 栏目:
-
图形与图像
- 出版日期:
-
2022-01-10
文章信息/Info
- Title:
-
Comparison and Analysis of Image Source IdentificationMethod Based on Neural Network
- 文章编号:
-
1673-629X(2022)01-0111-06
- 作者:
-
孟 旭; 孟 坤
-
北京信息科技大学 计算机学院,北京 100101
- Author(s):
-
MENG Xu; MENG Kun
-
School of Computer Science,Beijing Information Science and Technology University,Beijing 100101,China
-
- 关键词:
-
图像来源识别; 噪声提取; 神经网络; 特征提取; 传感器模式噪声
- Keywords:
-
image source identification; noise extraction; neural network; feature extraction; sensor pattern noise
- 分类号:
-
TP301
- DOI:
-
10. 3969 / j. issn. 1673-629X. 2022. 01. 019
- 摘要:
-
随着手机等便携式智能电子设备的普及,图像已成为最重要的信息载体之一,在新闻、社交及司法等领域发挥着重要作用。 在享用电子图像带来便捷性的同时, 图像处理工具给不法分子通过篡改电子图像实施诈骗等犯罪活动提供了可能,识别图像来源、辨别图像真伪已成为遏制和惩罚此类犯罪活动的重要技术手段。 该文讨论了神经网络在图像源识别中的应用方法,分别将原始图像和图像噪声作为模型输入数据,比较分析了神经网络的分类效果。 从依赖数据属性、数据预处理方法以及应用模式等方面进行了实验。 通过对实验结果进行分析,发现提取有代表性的图像块以及使用平滑的图像进行实验更有利于图像来源的识别。 分别采用笔者建立的数据集(10 个相机) 和 vision 数据集(35 个相机) 作为分析数据集,图像来源分类的实验结果表明相对于简单估计相机传感器模式噪声的方法准确率提升了 35% ,图像来源判断的实验结果准确率达到了 95% 。
- Abstract:
-
With the popularization of portable smart electronic devices such as mobile phones, images have become one of the mostimportant information carriers, playing an important role in news,social and judicial fields. While enjoying the convenience of electronicimages,image processing tools make it possible? ? ?for criminals to commit fraud and other criminal activities by tampering with electronicimages. Identifying the source of the image and distinguishing the authenticity of the image has become an important technical means todeter and punish such criminal activities. We discuss the application method of neural network in image source identification, andcompare and analyze the classification effect of neural network for the original image and image noise as model input data. Experimentsare carried out in terms of dependent data attributes,data preprocessing methods,and application modes. According to the analysis of theexperimental results,the extraction of overlapping image blocks and the use of smooth images for experiments are more conducive to theidentification of image sources. By using our own dataset (10 cameras) and vision dataset (35 cameras) as the analysis data sets,the experimental results of image source classification show that the accuracy of the method of simply estimating the camera sensor pattern noiseis improved by 35% . The accuracy of the experimental results of image source judgment reached 95% .
更新日期/Last Update:
2022-01-10