利用傅里叶变换提取图像纹理特征新方法

A new method for extracting image texture features by Fourier transform

研究发现,由图像傅里叶周向谱传统算法得到的频谱分布不能够真正反映其频率特性。因此,根据傅里叶变换的共轭对称性,提出了更具有一般性的长方环傅里叶周向谱能量百分比新算法。该算法均匀地把图像功率谱分成20个等间距同心长方环,计算每一个长方环内功率谱能量占总能量的比值作为图像频率分布特征。实验证明,新算法能更好地反映具有一般性的不同频率图像的纹理特征。在对作物缺乏营养元素诊断识别研究中,新算法提取的特征有效性远远高于传统算法,使识别的准确率达到82%以上。

It is discovered that the frequency spectrum distribution obtained by traditional algorithm of Fourier circular spectrum cannot really reflect its frequency behavior. So a new algorithm with rectangle loops Fourier spectral energy percentage is proposed by analyzing the conjugate symmetry of Fourier transform. The image power spectrum is divided into 20 equidistant concentric rectangle loops of the frequency domain by the algorithm. The ratio of the power spectrum energy within each rectangle loops making up the total energy will be calculated and the obtained value will be used as the feature of image frequency distribution. The experiments demonstrate that the new algorithm can reflect the texture features of the image with different frequencies better. In diagnosing plant disease for nutrients deficiency, the effectiveness of the features extracted by the new algorithm is far higher than that of the traditional algorithm. In this way the identification accuracy rate has attained more than 82%.