提出一种面向彩色图像的尺度和旋转不变性特征提取方法，并在真实的场景识别中进行了应用。该方法是先对给定彩色图像的各组成平面分别进行Radon变换，然后对得到的Radon变换系数矩阵进行尺度不变性处理，接着对处理后的Radon变换系数矩阵用频率B样条小波进行1维小波变换，在所得到的脊波系数矩阵中计算均值和方差的同时，采用线性回归模型提取在不同的颜色组成平面下所有频率子波段之间的关系属性，最后将得到的特征进行旋转不变性处理，从而得到所提出的尺度和旋转不变性特征。在3个数据库上进行了实验，结果表明本文方法可靠有效。

A scale and rotation invariant feature extraction method based on color images is proposed and been applied for the recognition of real-world scenes. Radon transform is firstly applied in the three component planes of the given color image and then the scale invariance operation is implemented in Radon transform coefficient matrixes. One-dimensional wavelet transform with the frequency B-spline wavelet is used in Radon transform coefficient matrix. The mean and variance are calculated in ridgelet coefficient matrix, and the relationship is extracted for all the frequency sub-bands under different color component plane using linear regression model. Finally rotation invariance operation is conducted for obtained features. Therefore, the proposed scale and rotation invariant features are acquired. The experimental results using the three databases show that the proposed method is reliable and efficient.