一种新的多尺度旋转不变性纹理特征提取方法  

A new multiresolution and rotation invariant texture descriptors

作　　者：孙慧贤[1] 张玉华[1] 罗飞路[1] 

机构地区：[1]国防科技大学机电工程与自动化学院,湖南长沙410073

出　　处：《光电子．激光》2010年第3期448-451,共4页Journal of Optoelectronics·Laser

基　　金："十一五"国防预研资助项目(51317030106)

摘　　要：提出一种基于局部Walsh谱(LWS)的多尺度旋转不变性纹理特征提取方法。首先通过比较每个像素点与邻近点的灰度值生成局部二值序列,然后计算局部二值序列的离散Walsh-Hadamard变换(DWT)的功率谱,最后采用功率谱的各谱点值构成特征直方图描述纹理特征。通过选择不同半径和采样点的局部二值序列可以得到不同尺度下的纹理特征,利用DWT功率谱的循环移位不变性可实现纹理特征的旋转不变性。纹理分类实验结果表明:与灰度共生矩阵(GLCM)、Gabor滤波器组等纹理特征相比,LWS在纹理鉴别能力和计算时间上具有较明显优势;与局部二值模式(LBP)相比,LWS在纹理分类准确率和旋转不变性方面均优于LBD。The sequency characteristics of local binary sequence is used to describe the gray variation of image neighbourhoods,and a new multiresolution and rotation invariant texture descriptor is proposed based on the local Walsh spectrum（LWS）.The characteristic histogram composed of LWS is used to describe the texture feature.The spectra of DWT on two texture classification experiments are implemented based on Brodatz and Outex texture database respectively.Comparing with Gray Level Co-occurrence Matrix（GLCM） and Gabor Filter Banks,LWS has the best texture discrimination performance and the least computational complexity.Comparing with Local Binary Pattern（LBP）,in the circumstances of same local neighbourhoods,the texture discrimination performances of LWS are better than that of LBP.

关 键 词：纹理分析 局部Walsh谱(LWS) 旋转不变性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]