支持向量机耦合韦伯局部描述符的图像伪造检测算法

摘要

针对当前图像伪造检测准确率低、适应性弱等问题，提出支持向量机耦合多分辨率韦伯局部描述符(WLD-multi-resolution Weber local descriptor)的图像伪造检测技术。将图像转换为 YCrCb模型，以亮度和色度的形式保存图像，利用人眼对色度的敏感性弱于亮度的特点，根据色度分量提取图像特征；根据当前像素的差分激励和梯度方向，构造 WLD计算模型，判断图像的灰度值变化和纹理变化信息；分析优化 WLD参数，采用像素点的差分激励和梯度方向在不同分辨率下形成 WLD直方图；引入支持向量机(SVM)分类算法对提取的伪造图像进行分类学习。在大量图像数据库进行伪造检测的实验结果表明，多分辨率 WLD相对单分辨率 WLD具有更好的检测效果，在拼接伪造图像和复制-移动伪造图像的检测中，多分辨率 WLD都有较好的检测性能。%In view that the current image forgery detection shows low accuracy and weak adaptability,a multi-resolution Weber local descriptors (WLD)image forgery detection technology was proposed.The image was converted into YCrCb model,and the image was saved in luminance and chrominance form.The characteristic that human eyes are less sensitive to chromaticity than brightness was used,according to the chrominance components of image feature extraction,the computing model of WLD was structured.Difference stimulation and gradient direction of the current pixel were calculated by WLD to determine the gray value of the image variation and texture change information.WLD parameters were analyzed and optimized,and the difference stimu-lus and gradient direction of pixel in different resolution were utilized to form WLD histogram.The support vector machine (SVM)classification algorithm was introduced to classify the forged image.The forgery detection experiments on large image database show that the detecting effects of multi-resolution WLD are better than single-resolution WLD,multi-resolution WLD has better detection performance on splicing forgery images and copy-move forgery images.