随着智能手机和社交媒体的普及,图像隐私引起的公共安全问题受到了广泛关注.犯罪分子可利用图像信息来侵犯他人隐私,从而进行恶意跟踪和诈骗勒索等活动.针对该问题,本研究从抑制图像源相机识别能力的角度提出了一种新的光响应非均匀性(photo response non uniformity,PRNU)匿名算法,以降低图像被恶意使用的风险.算法首先对图像进行分块,并将离散余弦变换(discrete cosine transform,DCT)和维纳滤波相结合,对图像高频分量进行滤波处理.通过实验发现,迭代维纳滤波可以显著提高匿名算法的效率.将算法在MICHE-I数据集和UBIPR虹膜数据集中进行测试,结果表明,算法能够保证图像质量不受显著影响,并且视觉质量因子平均提高3 dB,有效地抑制了图像溯源能力,保留图像生物特征,并达到了较好的匿名效果,保证了大众隐私安全.
Abstract
In the age of ubiquitous smartphones and the proliferation of social media platforms,the importance of image privacy has never been more prominent.The misuse of images by malicious actors has become a pressing concern,which has led to numerous illegal activities such as fraud,stalking,and extortion,and subsequently produced harmful consequences for our society.The primary objective of this study is to devise an efficient algorithm to tackle the predicaments stemming from the compromise of image privacy.We aim to combat these issues by delving into the internal information within images and mitigating the potential harm caused by image privacy breaches.The research primarily focuses on curtailing the identifiability of image source cameras,thereby reducing the potential for malicious image exploitation.Given the prevailing challenges associated with existing PRNU analysis techniques,including diminished quality and low time efficiency,this paper introduces a novel PRNU anonymization algorithm.Initially,the algorithm segments the image and employs a combination of discrete cosine transform(DCT)and Wiener filters to filter the high-frequency components within the image.Experimental results demonstrate that the iterative Wiener filter significantly enhances the efficiency of the anonymization algorithm.The proposed algorithm has been extensively tested on the MICHE-I dataset and the UBIPR iris dataset.The experimental outcomes reveal that the algorithm effectively preserves image privacy while maintaining minimal perceptible impact on image quality.The average improvement in visual quality factor is approximately 3 dB,and it successfully suppresses the traceability of the images.Furthermore,the biometric features of the original images remain intact,further enhancing the efficacy of anonymization.This research is of great significance to the fields of digital forensics and image analysis.The algorithm not only aids in safeguarding the privacy and security of the general public but also holds substantial promise in areas such as forensic investigations and criminal case resolutions.
维普
万方数据