基于方向—频率分解的纹理特征提取方法研究
Research of Texture Feature Extraction Method Based on Orientation-Frequency Decomposition
王超 1郭恒光2
作者信息
- 1. 上海东湖机械厂,上海 200439
- 2. 海军航空大学岸防兵学院,山东烟台 264001
- 折叠
摘要
为了解决纹理图像发生旋转时的识别问题,充分利用纹理在不同方向具有不同频率成分这一特性,结合Log-polar变换、Radon变换和经验模态分解,提出基于方向-频率分解的纹理特征提取方法.首先对图像进行Log-polar变换和Radon变换,实现图像的方向分解,得到Log-polar谱和Radon谱;然后对各个方向变换谱分量进行经验模态分解,实现图像在各个方向上的频率分解;最后对得到的各固有模态函数进行Hilbert变换,得到其包络和瞬时频率,对包络和瞬时频率的均值和方差进行傅里叶变换,得到具有旋转不变性的纹理特征.实验结果表明,提出的方法用于纹理图像识别时准确率高,抗噪声能力强.
Abstract
In order to solve the recognition problem when texture images undergo rotation,a texture feature extraction method based on direction frequency decomposition is proposed by fully utilizing the characteristic that textures have different frequency components in different directions,combined with Log polar transform,Radon transform,and empirical mode decomposition.Firstly,perform Log polar and Radon transformations on the image to achieve directional decomposition and obtain Log polar and Radon spectra;Then,empirical mode decomposition is performed on the spectral components of each direction transformation to achieve frequency decomposition of the image in various directions;Finally,Hilbert transform is applied to the obtained intrinsic mode functions to obtain their envelope and instantaneous frequency.Fourier transform is applied to the mean and variance of the envelope and instantaneous frequency to obtain texture features with rotational invariance.The experimental results show that the proposed method has high accuracy and strong noise resistance when used for texture image recognition.
关键词
纹理特征/方向-频率分解/Log-polar变换/Radon变换/经验模态分解Key words
texture feature/orientation-frequency decomposition/Log-polar transform/Radon transform/empirical mode decomposition引用本文复制引用
出版年
2024
NETL
NSTL
万方数据