【摘要】 该文结合多尺度技术与谱分析方法,提出了基于多尺度谱特征的图像分割方法,并将之用于SAR图像分割。该方法在多尺度框架内,提取每个像素在不同尺度下的局部谱特征(AR模型参数),并组合各尺度的谱特征为一多尺度谱特征向量,作为该像素的分类特征,利用一基于二元假设检验的分类器对该像素分类。与单一尺度的谱特征分割方法相比,多尺度谱特征分割保留了算法简单的优点的同时,在小窗口情况下,仍能给出较平滑的分割结果,从而减小了计算复杂度。更多还原

【Abstract】 A new segmentation method-multiscale spectral based segmentation,which combines multiscale technology with spectral analysis,is proposed,and then used for SAR image segmentation.In multiscale frame ,we get each pixel’s local spectral characteristics at different scales and combine them as the pixel’s multiscale spectral characteristic,and then use it to classify the pixel by a classifier based on binary hypothesis test.Compared with single scale spectral segmentation,while preserving the simplicity of the algorithm,the new method achieves more smooth segmentation on small window condition.And so,the computational complexity is reduced.更多还原