噪声方差和纹理复杂度分析的源相机识别

摘要

In allusion to source camera identification, a new method was proposed with noise variance and texture complexity analysis. Concerning Color Filter Array (CFA) interpolation and wavelet-based denoising, the insufficiency of traditional methods for extracting pattern noise was discussed first. Afterwards, the discussion was focused on the new extraction method according to noise variance and removing complex textured areas using fuzzy clustering. Experiments show that the extracted pattern noise not only reflects camera' s pattern noise well, but also improves the average accuracy with 6.3 percent for identifying images from three different camera models.%针对源相机识别问题,提出了一种利用噪声方差和纹理复杂度分析的源相机识别新方法.首先围绕彩色滤波阵列(CFA)插值和小波去噪,讨论了传统的模式噪声提取方法的不足,接着重点讨论利用噪声方差以及根据模糊聚类去除高纹理复杂区域进行模式噪声提取的新方法.实验表明所提取的模式噪声不仅能更好地反映数码相机的模式噪声特性,而且对来自三种不同相机的照片的平均识别率提高了近6.3％.