高空间分辨率遥感影像小波域分形纹理特征提取

融合小波多尺度分析方法及分形纹理提取方法在遥感影像信息提取方面的优势，提出高分辨率遥感影像小波域分形纹理特征计算方法，以获取地物多尺度分形纹理属性，为遥感影像地类识别提供更好的标识。首先对遥感影像进行小波多尺度分解，进而基于DBC、多重分形纹理计算方法在各个分解层上提取地物纹理特征，通过比较分析，从中选取更为有效的小波域分形纹理特征。基于该方法，利用福州市高空间分辨率QuickBird遥感影像进行试验，并对QuickBird影像进行三级小波分解及纹理提取，结果表明：小波第一、第二分解层粗影像（CA1、 CA2）及三方向平均细节影像（L1、 L2）的DBC空隙特征及多重分形分维数结果作为最终甄选的小波域分形纹理特征更为合适。

Extracting Wavelet-Domain Fractal Texture of High-Resolution Remotely Sensed Imagery

For high-resolution remote sensing images of abundant texture information and multi-scale features, an approach of wavelet-domain fractal texture extraction was proposed in this research based on the fractal texture extraction and wavelet multi-scale analysis methods. The multi-scale textures could play an important role on automatic recognition of ground objects in images. First, wavelet decomposition of the image was made. Then, the fractal texture of each decomposition image was extracted based on DBC and multi-fractal methods. Finally, the texture features was analyzed and compared for all decomposition images, and the more effective texture was selected for identification of objects. A demonstration was made with the high-resolution QuickBird image of Fuzhou, and the conclusion was that the DBC cap and multi-fractal features of coarse images （ CA1 , CA2 ） and mean fine images on three directions （ L1 , L2 ） of the first and second decomposed layers in three decomposed ones of QuickBird image, were the most effective wavelet-domain fractal information. This research provides an effective method of multi-scale fractal texture extraction, and will improve the identification results of ground objects in high-resolution remote sensing images.