论文部分内容阅读
为了减少仅用分水岭变换而导致的过分割问题,本文提出利用小波变换的多尺度处理方式用于融合后多光谱QuickBird图像的分割。整个分割过程包括多尺度图像表示、图像分割、区域合并和结果映射等过程。首先,依据原始图像的大小确定分解尺度并用小波变换产生各波段的低尺度图像。采用相位一致模型提取各近似系数的梯度,并逐尺度地融合各梯度图。分析不同尺度下的不同地物的局部梯度方差,以选择最佳的小波分解尺度。然后,通过移动阈值与扩展最小变换,利用多层次标记提取方法标记均质区域。进而,在梯度重建的基础上利用标记分水岭变换得到分割图像。其次,采取空间相邻关系、面积、光谱与纹理等多约束策略,以搜索最小合并代价的方式合并最初分割区域中的邻接区域对。最后,修改细节子图并进行小波逆变换将最初分割结果投影到更高尺度图像,同时处理边界上的像元以保持区域边界直至原始图像。实验结果表明本文方法不仅能够用于高分辨率多光谱遥感图像的分割,而且缓解了过分割问题且取得了较准确的分割效果。
In order to reduce the over-segmentation problem caused by only the watershed transformation, this paper proposes a multi-scale processing method using wavelet transform for the segmentation of the fused Multispectral QuickBird image. The whole segmentation process includes multi-scale image representation, image segmentation, region merging and result mapping. Firstly, the decomposition scale is determined according to the size of the original image and the low-scale image of each band is generated by wavelet transform. The phase-consistent model is used to extract the gradient of each approximation coefficient, and the gradient maps are fused on a scale-by-scale basis. The local gradient variances of different features under different scales are analyzed to select the best wavelet decomposition scale. Then, by moving the threshold and expanding the minimum transform, the homogeneous region is marked by the multi-level mark extraction method. Then, based on the gradient reconstruction, the segmentation watershed transform is used to obtain the segmentation image. Secondly, taking the multi-constraint strategy such as spatial neighbor relationship, area, spectrum and texture, the adjoining pairs in the initial segmentation region are merged in the way of searching the minimum merge cost. Finally, the detail subgraph is modified and the inverse wavelet transform is used to project the original segmentation result to a higher-resolution image while processing the pixels on the boundary to keep the boundary of the region up to the original image. Experimental results show that this method can not only be used for high resolution multi-spectral remote sensing image segmentation, but also alleviate the over-segmentation problem and achieve a more accurate segmentation results.