基于二元树复小波和SVD不变性的纹理图像分类算法

针对双树复小波变换(DT-CWT)用于图像的纹理特征提取时,不具有旋转和尺度不变性的局限,提出了一种基于DT-CWT和SVD的纹理分类算法.该算法首先利用DT-CWT从图像中提取出纹理特征,然后对纹理特征进行奇异值分解获得具有旋转和尺度不变性的特征向量,采用BP神经网络作为分类器,并提出改进的BP算法训练网络,使得网络很快找到全局最优解.将本方法与其他的分类算法进行比较,实验结果表明,本算法具有较高的分类正确率.

Sort Algorithm for Invariance Texture Image Based on DT-CWT and SVD

In viewing of the disadvantage of dual-tree complex wavelet（DT-CWT） which has no rotation and scale invariance when it is used to extract texture feature of images,a texture sort algorithm based on DT-CWT and singular value decomposition（SVD） was proposed in this paper.This algorithm extracted the texture feature from the images using DT-CWT.Then SVD was employed to obtain invariant feature vector under translation,rotation and scaling.BP neural network was adopted as a classifier.An improved BP algorithm was also proposed to train the network to find entire optimization solution rapidly.Experimental results show that the proposed algorithm has better sort accuracy than other classification algorithms.