论文部分内容阅读
SIFT算法描述子均匀量化局部区域的位置和梯度角,难以有效解决描述子局部几何失真的问题。该文改进SIFT描述子的汇聚策略,将描述区域不规则分割(位置量化)与梯度角方向柱递减(方向量化)相结合,提高特征描述子抵抗局部几何失真的能力。将本文算法与原SIFT描述子进行比较,使用不同条件的遥感影像对测试,从匹配点对数量、时间耗损和稳定性等方面进行对比试验,验证了不规则量化描述子有较强的独特性和鲁棒性。
It is difficult to effectively solve the problem of describing local sub-geometrical distortions by using the SIFT algorithm to uniformly quantize the location and gradient angles of local regions. In this paper, the convergence strategy of SIFT descriptor is improved, and the combination of region irregular segmentation (position quantization) and gradient angle directional column decrement (directional quantization) is described to improve the ability of feature descriptors to resist local geometric distortions. Comparing the proposed algorithm with the original SIFT descriptor, using different conditions of the remote sensing images to test, comparing the number of matching points, time loss and stability, we verify that the irregular quantization descriptor has strong uniqueness And robustness.