【摘要】 视觉显著性检测是一类有效的遥感图像目标提取方法。在计算过程中,多数现有显著性检测模型都需要使用图像的色彩分量,针对灰度图像的算法较为匮乏。此外,变换域算法相较于空间域算法具有计算复杂度低和物理意义明确的优势,但现有模型随输入图像提取目标性能差异较大。针对上述问题,本文基于多尺度平稳小波(SWT)分解,提出了一种灰度遥感图像目标提取方法。通过对输入图像进行不同尺度的SWT分解和重构,得到一系列特征图。得到的特征图通过二维熵准则进行加权融合,得到显著性图。最后对显著性图进行大津(OTSU)算法分割和形态学闭处理提取出目标。实验证明,相较于现有变换域显著性模型,本方法在主观视觉效果和客观评价标准上都具备一定优势。更多还原

【Abstract】 Visual saliency detection is an effective approach for target detection in remote sensing images.Most existing saliency detection models use the color components of images,and algorithms for gray-scale images are relatively scarce. Additionally,methods in transform domain are superior to spatial domain methods in lower computational complexity and explicit physical meaning. However,their performance varies with different input images. To solve this problem,we propose an object extraction method for grayscale remote sensing images based on stationary wavelet( SWT) decomposition. By applying multi-level SWT decomposition and reconstruction to input images,a series of feature maps are obtained. Then,the obtained feature maps are weighted by two-dimension entropy and fused as a saliency map. Finally,the object is extracted by employing OTSU segmentation and morphological close processing. Experimental results show that the proposed scheme has better performance in both subjective vision quality and objective evaluation compared with existing transform saliency models.更多还原