采用HOG特征的下视景象匹配算法

Approach on scene matching based on histograms of oriented gradients

红外实时图像和可见光参考图像的差异给下视景象匹配技术带来挑战,如何寻找并提取红外实时图像与可见光基准图像间稳定和可靠的共性特征是进行特征匹配的重要研究课题。在分析红外图像和可见光基准图像成像机制基础上,提出了以HOG特征作为匹配的特征、以相关系数作为相似性测度的匹配算法,并研究了梯度方向范围、梯度方向划分、标准化等计算HOG特征相关参数对于匹配性能的影响,给出了性能较优的一组参数。实验表明:该匹配算法与互相关和Hausdorff距离法相比具有更高的正确匹配率。

The great differences between infrared images and visual images bring challenge to scene matching.How to find the common features which exist synchronously between infrared images and visual images and how to extract them are important issues in scene matching based on features.By analyzing the imaging mechanism of infrared images and visual images firstly,a novel matching method for scene matching was proposed based on histograms of oriented gradients（HOG） used as the matching feature and the correlation coefficient used as the similar measure.The selection of the parameters on the proposed algorithm,such as the range of gradient orientation,number of bin on gradient orientation,block normalization etc.was discussed and an appropriate parameter values were presented for excellent matching performance.Experiments showed that the correct matching probably of the proposed method was higher than that of cross-correlation matching algorithm and Hausdorff distance matching algorithm.