基于小波包多尺度变换的旋转伸缩不变特征提取

Feature Extraction of Rotation and Scale Invariant Based on Wavelet Packet Multi-Scale Transform

离散小波变换定位频域和空间时,由于基本小波变换对旋转和伸缩敏感,故变换前应预处理图像,步骤为:直角坐标系到对数极坐标的转换,自适应行移不变性小波包转换,最佳树中各子带小波包系数能量特征向量的不变特征提取。

When the frequency domain and space was defined with discrete wavelet transform, basic wavelet transform is sensitive to rotation and scale invariant. Before wavelet transform, the image was preprocessed. Rectangular coordinate system was transformed into logarithm-polar coordinate to get partly rotation and scale invariant at row shift. The rotation and scale invariant feature of image was extracted with row shift invariant wavelet packet transform.