多特征的灰度反转和旋转不变直方图

Multi-feature Grayscale-inversion and Rotation Invariant Histogram

针对图像发生反向灰度变化时,传统局部二值模式及其派生的纹理描述符分类性能会明显下降问题,提出了一种灰度反转和旋转不变的直方图方法。该方法通过高斯导数滤波获得更加细微的图像微分结构,再构建具有旋转不变性的多种互补特征,最后利用主导思想对多种互补特征进行联合编码,获得具有灰度反转和旋转不变性的特征直方图。使用Outex、CUReT和KTH-TIPS等3个纹理图片数据库验证所提出方法的分类性能。实验结果表明:在旋转和灰度反转情况下,提出的方法能够显著提高分类性能,有效改善灰度反转问题,并对图像的旋转变化具有一定的鲁棒性;从算法复杂度来看,该方法取最佳尺度及特征时,运行时间适中,整体分类精度高。

Regarding the problem that classification performance of the traditional local binary pattern(LBP)and LBP-derived texture descriptors decrease dramatically when images suffer from inverse grayscale changes,a grayscale-inversion and rotation invariant histogram is proposed.Firstly,the differential structure of the image is obtained by Gaussian derivative filtering.Then,multiple complementary features with rotation invariance are constructed.Finally,the dominant idea is used to jointly encode multiple complementary features and rotation and grayscale-inversion invariant feature histograms can be obtained.Three texture image databases i.e.,Outex,CUReT and KTH-TIPS,are used to verify the classification performance of the method.Experimental results show that the proposed method can significantly improve classification performance under image rotation and grayscale inversion.The complexity of the algorithm is analyzed.When the method takes the best scale and features,the running time is moderate,and the overall classification accuracy is the best.