多尺度梯度域可见光与红外热图像融合方法研究

Research of multi-scale gradient domain visible and thermal image fusion method

针对红外热像图缺乏细节信息而导致电气设备热故障定位困难的问题,提出一种基于结构张量的多尺度梯度域可见光与红外热图像融合算法。采用多尺度变换与温升区域特征图构造梯度权重,结合带有梯度权重的结构张量与变分技术重建图像,并运用透明度法进一步融入可见光图像的细节信息,同时保留了源图像的温升区域与背景信息。实验表明,该算法得到的融合图像细节丰富、冗余信息少;主观视觉效果与客观评价指标均优于混合图像法、小波变换法、IHS变换法以及快速红外图像融合算法(FIIF),为电气设备热故障准确定位提供了一种方法。

Considering the difficulty to find the thermal fault position of electrical equipment resulted from the lack of detail information of the thermal images,this paper proposed a multi-scale gradient domain visible and thermal image fusion algorithm based on structure tensor. Firstly,the algorithm computed the gradient weights by multi-scale structure and the regional characteristic of high temperature images. Then it reconstructed the images through the structure tensor with weights and variation techniques. Finally,it obtained the fusion images by transparency method,which maintained the high temperature and the background information of the source images. The experimental results clearly show that the fusion images of proposed algorithm contain abundant details and less redundant information. In addition,the proposed algorithm outperforms the conventional fusion methods both in subjective and objective perspectives,providing a method for fault monitoring of electrical equipment.