The extraction of SPN (sensor pattern noise) is a key procedure for images source identification. But the SPN extracted by existing methods are susceptible to the interference of texture and scene in images. For these reasons

a novel method is proposed for SPN extraction based on combined filtering. Firstly

images are analyzed with an orthogonal wavelet transform

an edge preserving bilateral filter and a local adaptive minimum mean squared error filter for the approximation and detail subband respectively. Then we denoised with bilateral filter in spatial domain. Finally

the classifier based on correlation detection principle is constructed with 9 cameras reference SPN

and accurate identification for SPN has been achieved. Then we analyzed the results of 3 kinds of typical size. The experimental results indicate that the proposed method effectively reduces the interference due to scene. A 79.32% accuracy has been achieved even the size of detected images are 256×256 pixels.