数字图像与数码相机噪声相关性的偏态分布

Asymmetrical distribution of noise correlation between digital image and digital camera

为了提高利用噪声相关性鉴别数字图像真伪的正确性,探索了描述噪声相关性偏态分布的最佳模型。首先,通过理论分析和大量实验验证了在(0,1]区间beta分布、gamma分布和对数正态分布可用于描述噪声相关性的偏态分布。然后,利用这3种分布的概率密度函数模拟实际的噪声相关性偏态分布的概率密度函数曲线,概率密度函数曲线的特征和最小错误率的大小说明了采用对数正态分布描述噪声相关性偏态分布的效果最佳,从而提出用对数正态分布描述噪声相关性偏态分布的模型。实验结果表明,与采用广义chi平方分布的模型相比,采用该模型可使最小错误率降低60%以上,证明了采用正确的模型描述噪声相关性偏态分布是降低鉴别错误率的有效途径。

The best model to describe the asymmetrical distribution of noise correlation was explored to improve the accuracy of identifying digital image authenticity by noise correlation. By theoretical analysis and experiments, it was indicated that the beta distribution, gamma distribution and the logarithmic normal distribution could be used to describe the asymmetrical distribution of noise correlation within （0,1] . Then, three kinds of probability density functions were used to simulate the probability density function curves of the actual asymmetrical distribution of noise correlation. On the basis of the curve modality of probability density function and the value of the least false rate,it was explained that using logarithmic normal distribution to describe the asymmetrical distribution of noise correlation could obtain the best result. Thus a model by using the logarithmic normal distribution to describe the asymmetrical distribution of noise correlation was proposed. Experiment results show that this model can reduce the least false rate over 60% as compare with the model of using generalized chi square distribution, which proves that using a correct model to describe the asymmetrical distribution of noise correlation is an effective approach to reduce the false rate.