基于噪声方差估计的伪造图像盲检测方法

Blind image forgery detection method based on noise variance estimation

针对不同来源合成伪造数字图像提出了一种盲检测方法,不同数字图像背景噪声存在差异,因而伪造图像区域噪声方差不同。从待测图像小波分解后的高频子带中去除相应边缘区域的高频干扰,改进了噪声方差估计算法,并对所得噪声图像进行分块处理估计每一个分块的噪声方差,将方差相近的块进行融合,最后比较图像中纹理接近的同质区域,找出方差异常的位置。通过实验研究了方差估计精度,对不同来源的伪造图像进行了检测,结果表明算法提高了图像噪声方差的估计精度,在图像纹理接近的同质区域中可以定位图像的伪造区域。

This paper developed a new blind detection method on the synthesis of forgery digital images from different sources. Different digital image background noise variance was different. First of all, it decomposed the image by wavelet decomposition, and removed the high frequency interference of the corresponding region of the image from the HH sub band, improved the noise variance estimation algorithm. And then it divided the noise image into blocks and estimated the noise variance of each sub block. In the homogeneous region of texture, the fusion of the block with the closed variance was carried out to find out the region of the abnormal variance. The experimental results show that the proposed algorithm can improve the estimation accuracy of image noise variance, and locate the forgery image regions in homogeneous areas.