一种用于电力设备图像配准的改进SIFT方法

An Improved SIFT Method for Image Registration of Power Equipment

针对传统图像配准的尺度不变特征转换算法存在实时性差、误匹配率高等问题,提出尺度不变特征转换算法的改进方法.对尺度不变特征转换算法关键点的特征向量和梯度进行归一化来去除光照变化的影响后,再对某些方向过大的梯度值设置门限值并再次归一化,从而提高特征的鉴别性.试验表明改进后的配准算法提高了特征匹配的精确度,配准的正确率得到了提升.

Power equipment failure will affect the safe and stable operation of power grids.Image recognition of power equipment is widely used in the maintenance of power equipment.The accuracy of image recognition can be improved based on registration of infrared and visible images.Traditional SIFT algorithm for image registration is fraught with problems of poor real-time performance and high error matching rate.To address those problems,an improved SIFT algorithm is proposed.After normalizing the feature vectors and gradients of sift key points to remove the influence of illumination changes,this paper sets threshold values for excessive gradient values in some directions and normalized again,so as to improve the discrimination of features.Experiments show that the improved registration algorithm increases the accuracy of feature matching and registration.