CAJ | 학술논문

在局部二进制模型和多分辨率分析的基础上，提出一种基于局部频域分析的多尺度纹理特征提取方法。对纹理图像采用局部圆形邻域作为局部模板进行局部频率变换，对各频段图像实现多分辨率分析，提取多尺度下的特征向量，结合SVM算法进行分类实验。结果证明，该方法在继承了LBP对光照因素具有的稳定性的同时，具有旋转不变性特征；利用NSCT在多尺度下进行的纹理特征提取也满足尺度不变性的需要，与其他方法相比提高了分类准确率。
재국부이진제모형화다분변솔분석적기출상，제출일충기우국부빈역분석적다척도문리특정제취방법。대문리도상채용국부원형린역작위국부모판진행국부빈솔변환，대각빈단도상실현다분변솔분석，제취다척도하적특정향량，결합SVM산법진행분류실험。결과증명，해방법재계승료LBP대광조인소구유적은정성적동시，구유선전불변성특정；이용NSCT재다척도하진행적문리특정제취야만족척도불변성적수요，여기타방법상비제고료분류준학솔。
Based on the analysis of local binary pattern and multi-resolution, this paper presented a multi-scale texture classification method based on local frequency operator. The local frequency conversion was performed in the original image, and each sub-band image had multi-resolution analysis and multi-scale extraction of feature vectors. Classification experiments combining SVM algorithm was then realized. Experimental results show that this method inherits stability of LBP to light factors, while maintaining features to be invariant to rotation. The texture feature extraction meets the needs of scale invariance by using NSCT, and enhances the classification accuracy compared with other methods.