摘要
针对红外/可见光异源图像匹配,提出基于无迹变换的KL散度异源图像匹配方法。首先,分别提取异源图像Sobel边缘特征点,并校正红外特征点集;然后,分别构建两特征点集的高斯混合模型,采用无迹变换法求解两高斯混合模型的KL散度,对红外特征点集的高斯混合模型进行相似变换,寻找到使KL散度最小时的变换矩阵Tmin,此即两特征点集精确匹配时的变换矩阵。实验结果表明:提出的算法在噪声和出格点较多情形下仍能正确匹配,且能快速收敛。
A method for infrared/optical multi-sensor image matching based on Kullback-Leibler (KL) divergence of unscented transform was proposed. Firstly, we extracted Sobel edge points sets from the multi-sensor image respectively, and rectificated forward-looking infrared image point set; and then, we established Gaussian mixture models (GMM) on the two point sets respectively, and calculated the Kullback-Leibler divergence between the two GMM by unscented transform, and used derivative-free method to minimize the KL divergence, which the corresponding transformation matrix Tmin is the matching matrix of the two point set ; Experimental results show that the proposed algorithm is robust to noise and outliers.
出处
《重庆理工大学学报(自然科学)》
CAS
2016年第8期137-142,共6页
Journal of Chongqing University of Technology:Natural Science
基金
国家自然科学基金资助项目(61401504)
中国博士后科学基金资助项目(2014M562562)