基于纹理免疫的JPEG预压缩图像降尺度因子检测

图像编辑工具的普及导致JPEG图像越来越容易被篡改，其中重采样操作方法通过几何变换能够使伪造的图像更加逼真，因此对JPEG图像进行重采样检测至关重要。传统的光谱分析法依据差分图像相邻极值的间隔遵循几何分布，以及在直方图上的峰值呈现周期性，通过峰值分析对下采样因子进行估计，但由于图像纹理统计特征的周期性干扰了直方图的提取，导致检测准确性较低。提出一种用于JPEG预压缩图像降尺度因子检测的纹理免疫块效应分析算法，利用快速导向滤波对图像进行预处理，并去除图像的纹理和噪声。使用Canny算子对滤波后的图像进行边缘检测，以减轻周期性边缘的干扰，对图像进行交叉差分，从而凸显块效应，提高JPEG图像降尺度因子估计的准确性。在此基础上，结合极大似然估计和谱分析得出降尺度因子的估计值，减小估计误差。实验结果表明，该算法能有效削弱图像纹理对重采样估计的影响，具有较强的纹理免疫能力。

Downscaling Factor Detection of Pre-JPEG Compressed Images Based on Texture-Immune

The popularity of image editing tools makes JPEG images increasingly vulnerable to tampering, and the resampling operation method makes forged images more realistic through geometric transformation.Resamplingof the detection of JPEG images is therefore crucial.The traditional spectral analysis method follows a geometric distribution according to the interval of adjacent extreme values of the difference image and presents periodic peaks on the histogram.Thedownsampling factor is estimated through peak analysis.The periodicity oftheimage texture statistical features interferes with the extraction of the histogram, resulting in low detection accuracy.A texture immune block effect analysis algorithm for the detection of the downscaling factor of the pre-JPEG compressed images is proposed.The image is pre-processed by fast-guided filtering, and the texture and image noise are removed.The Canny operator detects the edge of the filtered image to reduce the interference of periodic edges, and the cross difference highlights the block effect and improves the accuracy of the JPEG image downscaling factor estimation.On this basis, combined with maximum likelihood estimation and spectral analysis, the downscaling factor is estimated to reduce the estimation error. Experimental results show that the algorithm can effectively weaken the influence of image texture on resampling estimation and provide strong texture immunity.