光电工程, 2012, 39 (2): 109, 网络出版: 2012-02-20  

SPGD图像匹配的方法性能及参数选取优化策略

Performance and Parameters Selection Optimization Strategy of Stochastic Parallel Gradient Descent Image Matching Method
作者单位
1 国防科学技术大学光电科学与工程学院,长沙 410073
2 总装备部武汉军代局驻南阳地区军代室, 河南 南阳 473678
摘要
随机并行梯度下降 (SPGD)图像匹配方法是一种全新的图像匹配方法,该方法同时对所有的变形参数施加相互统计独立的随机扰动,并以相关系数随机梯度分量代替真实的梯度分量进行迭代运算,使之能够快速地得到最优参数估计。扰动幅度和增益系数选取是该方法亟需解决的关键性问题,目前能否对不同的变形情况选取相同的具有普遍适应性的预设参数,或是需要对不同的变形情况自适应更新参数以达到最优化匹配效果都还没有得到较好的回答。本文一方面探索该方法在不同预设条件下的极限匹配性能;另一方面初步总结出扰动幅度和增益系数对该方法匹配性能影响的规律,在此基础上研究了分级参数更新策略,使该问题得到进一步解决。
Abstract
Stochastic Parallel Gradient Descent (SPGD) image matching method is a new image matching method, which simultaneously imposes statistically independent random disturbances on all of the deformation parameters and use correlation coefficient stochastic gradient component to carry out an iterative operation instead of real gradient component, so that it can obtain quick optimum parameter estimation. The selection of the disturbance amplitudes and the gain coefficients is the key issue that needs to be solved in this method. Presently, the question that, if we could select the same default parameters of universal adaptability in different deformation situation, or if we should update parameters to achieve the optimal matching results, has not got a better answer yet. On the one hand, this method’s utmost matching performance is explored in different preset conditions; on the other hand, it elementarily summarizes the laws of the impact on this method’s matching performance by selection of different disturbance amplitudes and gain coefficients. On the basis of the above, the grading parameters updating strategy is studied, so that this issue has been further addressed.

李松洋, 伏思华, 尹世亮, 龙学军, 王三宏, 韩海涛. SPGD图像匹配的方法性能及参数选取优化策略[J]. 光电工程, 2012, 39(2): 109. LI Song-yang, FU Si-hua, YIN Shi-liang, LONG Xue-jun, WANG San-hong, HAN Hai-tao. Performance and Parameters Selection Optimization Strategy of Stochastic Parallel Gradient Descent Image Matching Method[J]. Opto-Electronic Engineering, 2012, 39(2): 109.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!