基于多尺度方向分解的纹理特征提取方法研究
投稿时间:2020-09-28  修订日期:2020-11-10  点此下载全文
引用本文:
摘要点击次数: 27
全文下载次数: 0
作者单位E-mail
郭恒光 海军航空大学岸防兵学院 guohengguang@126.com 
朱 默 海军航空大学保障部  
赵 亮 海军航空大学岸防兵学院  
中文摘要:为了有效提取图像的纹理特征,充分利用纹理的方向性以及纹理在不同方向具有不同频率成分这两个特性,提出了基于多尺度方向分解的纹理特征提取方法。首先根据Radon变换的方法检测纹理方向,Radon变换各角度投影向量方差的二阶导数最小值对应的投影角度即为纹理方向。然后根据得到的图像纹理方向信息,利用可控金字塔将图像沿纹理方向进行三个尺度的分解,得到纹理方向上图像的多尺度子带图像。最后以三个尺度子带图像的Legendre矩和Zernike矩作为图像的纹理特征。分别在Brodatz和VisTex数据集上进行实验验证,与其他方法的对比结果表明,采用多尺度分解的方法提取纹理,用于纹理图像识别时,识别准确率高,抗噪声能力强。
中文关键词:方向分解  可控金字塔变换  纹理特征
 
Texture Feature Extraction Method Based on Multiscale Orientation Decomposition
Abstract:In order to extract the texture feature of the image effectively, the texture feature extraction method based on multiscale orientation decomposition was proposed, which can make full use of the texture orientation and the property that the texture has different frequency component at different orientation. Firstly, the texture orientation was detected by the Radon transform, the projection angle corresponding to the minimum second derivative value of the projection vector variance of the Radon transform at different angle is the orientation of the texture. Secondly, according to the texture orientation information of image, the steerable pyramid transform was used to decompose the image multiscale along the texture orientation, and the multiscale subband of the image can be obtained at the texture orientation. At last, the Legendre moment and the Zernike moment of each scale subband image were used as the texture feature of the image. The proposed method was verified in the Brodatz and VisTex datasets. Compared with other methods, the results show that the proposed method can reach the higher correct classification rate, and has stronger anti-noise ability.
keywords:orientation decomposition  steerable pyramid transform  texture feature extraction
查看全文   查看/发表评论   下载pdf阅读器