Abstract: By applying the filter which is in the wavelet domain, a certain noise is extracted which can reflect the camera physical characteristic and it further serves as the key feature to detect the authenticity of digital image. The noise characteristics of suspicious region in the image are determined and aggregated by the generalized Gaussian classifier and BP neural network classifier, to detect forgeries in digital images. The experimental results show the approach provides a relative high accuracy on detecting digital images that are forged in various ways.

Key words: camera noise, wavelet domain filter, generalized Gaussian, Back Propagation（BP） neural network, authenticity of digital image detection

摘要： 针对数字图像的真伪鉴别问题，通过在小波域上构造的滤波器，提取反映相机本身物理特性的某种特定噪声，将其作为判断图像真伪的关键特征。在待测图像中选取出可疑区域，将其噪声特征通过广义高斯分类器以及BP神经网络分类器进行判断和融合，从而实现图像的真伪鉴别。实验结果表明，该方法对多种不同伪造方式的数字图像均具有较高的识别正确率。

关键词: 相机噪声, 小波域滤波器, 广义高斯, 反向传播（BP）神经网络, 图像真伪鉴别