【摘要】 针对灰度相关法鲁棒性差的缺点,提出一种以图像特征进行匹配的新方法。采用具有良好稳定性和旋转不变性的图像奇异值分解作为匹配特征,并根据奇异值的数值特点,采用规范化奇异值的Hausdorff距离作为特征匹配的相似性度量;通过变模板分级匹配的策略,在保证匹配精度的前提下,使匹配时间减少到约10%。在加入方差σ=35的高斯噪声,灰度变化25%的情况下对旋转变化的景像目标进行匹配实验,结果证实了本文算法的有效性和鲁棒性及快速性。更多还原

【Abstract】 Traditional scene matching method is susceptible to environmental noise. To overcome this problem, taking into account singular values’ stability and rotation-invariance a new scene matching method based on Singular Value Decomposition (SVD) is proposed. Based on singular value’s unique characteristics, it is further proposed to use Hausdorff distance as the similarity measure between images. In addition, a new coarse-to-fine hierarchical matching approach is proposed to significantly reduce the computation for 90%. The proposed algorithm is tested on various corrupted images with Gaussian noise (σ=35), intensity changes (25%) and rotation changes. Experimental results demonstrate the robustness and accuracy of the algorithm.更多还原