针对目前纹理识别算法对强剪切攻击识别率不高的问题,提出了一种基于非负矩阵分解(NMF)结合不变矩抗剪切攻击和涂抹攻击的纹理识别算法.该算法首先对训练纹理图像样本y进行非负矩阵分解得到基矩阵W分量和系数矩阵H分量,并将其进行存储,同时计算W分量的不变矩作为图像特征向量；对经过剪切的测试样本图像,通过局部未剪切部分V矩阵和相应的H矩阵来恢复W矩阵,然后计算其不变矩作为特征向量；最后用K近邻分类器对提取的特征向量进行分类.通过对50类纹理图像进行分类实验,结果表明:本算法对遭受剪切攻击后的纹理图像具有很强的鲁棒性,识别率可达100％,对于遭受脉冲噪声感染后的纹理图像的识别率也在85％左右.%Aiming at the problem that the existing texture recognition algorithm was insufficient to strong shearing and smearing attack, a novel texture recognition algorithm was proposed by non-negative matrix factorization (NMF) and invariant moments. Firstly, non-negative matrix factorization (NMF) was used to translate the training texture samples matrix V into the base matrix W and the corresponding coefficient matrix H, meanwhile, the corresponding coefficient matrix H was stored. Then, invariant moments of matrix W was calculated as feature vector. For shearing and cropping training samples image, matrix W could be restored by matrix V and corresponding matrix H of the existing non-shearing image. Then, invariant moments of matrix W was calculated as feature vector. Finally, K nearest neighbor classifier was used to classify the extracted feature vector. The experimental results showed that the proposed method has more strong robust for shearing and smearing texture images, and the correct classification results ( CCPS ) was 100%. Meanwhile,the CCPS was about 85% for texture image attacked by impulse noise.
展开▼
