基于NSCT域FAST角点检测的电气设备红外与可见光图像配准

Registration based on NSCT-domain FAST corner detection for infrared and visible images of electrical equipment

针对电气设备在线监测系统中红外与可见光图像配准问题,提出了基于非下采样轮廓波变换(Non-Subsampled Contourlet Transform,NSCT)域改进的加速分割检测特征(Features from Accelerated Segment Test,FAST)提取的图像配准方法。首先采用灰度均衡技术对原始图像进行图像增强;再通过一级NSCT变换得到红外与可见光的低频子带图像;然后对低频子带图像利用FAST角点检测以及局域强度不变特征描述符(Partially Intensity In-variant Feature Descriptor,PIIFD)得到特征描述点对;最后,利用最近邻距离比率法及随机抽样一致性算法(Random Sample Consensus,RANSAC)得到精匹配点对,进而计算仿射变换参数。实验结果表明,相比传统的PIIFD配准算法,该方法的特征点正确匹配率提高了2. 52%,配准均方根误差降低11. 58%。

Infrared and visible image registration has become a big challenge in online monitoring system for high voltage electric equipment.Aiming at this problem,an efficient registration method based on features from accelerated segment test(FAST)in non-subsample contourlet transform(NSCT)domain is proposed in this paper.Firstly,the two images are preprocessed by gray-level equalization.After that,NSCT decomposition is applied to obtain low-frequency subband images.And corner points are detected from the low-frequency images as interest points by FAST.Their descriptors are calculated by partially intensity invariant feature descriptor(PIIFD)and used to make rough matching through best bin distance ratio.Finally,random sample consensus(RANSAC)is further utilized to refine the matched interest point pairs and affine transformation parameters between infrared and visible images are determined.Objective evaluation on experimental results shows that,compared to the traditional registration method based on PIIFD,the correct matching ratio of the proposed registration method is improved by 2.52%and the root mean square error is reduced by 11.58%.