Thermal imaging is a useful imaging technique in many scenarios. It can capture the temperature distribution of scenes in the dark and see through sparse smoke and dust. However, some surfaces such as steel and glass with high reflectivity lead to a reflection problem in thermal imaging, while heavy mist and gases lead to the occlusion problem. We proposed an efficient algorithm to solve the occlusion problem in our earlier work. The reflection in thermal images causes errors in detection and temperature measurement. Therefore, the precise model and efficient algorithms to solve this problem are in high demand. In this paper, we mainly model the reflection problem in thermal imaging and propose an algorithm to deal with it. In our experiments, a thermal camera array is built to capture the thermal light-field images. We first separate a part of the reflection pixels from thermal images based on the depth information. After that, the thermal reflection is removed by optimizing a designed cost function. The experiment results show that our reflection removal method can separate the thermal reflection with high precision, retain the objects in the scene, and get better performance than existing methods.

中文翻译：

---

使用光场理论基于深度的热反射去除

热成像在许多情况下都是一种有用的成像技术。它可以在黑暗中捕捉场景的温度分布，透过稀疏的烟尘看清。但是，一些具有高反射率的表面（例如钢和玻璃）会导致热成像中出现反射问题，而重雾和气体会导致遮挡问题。我们在早期的工作中提出了一种有效的算法来解决遮挡问题。热图像中的反射会导致检测和温度测量出现错误。因此，迫切需要精确的模型和有效的算法来解决这个问题。在本文中，我们主要对热成像中的反射问题进行建模，并提出一种算法来处理它。在我们的实验中，构建了一个热像仪阵列来捕获热光场图像。我们首先根据深度信息从热图像中分离出一部分反射像素。之后，通过优化设计的成本函数来消除热反射。实验结果表明，我们的反射去除方法可以高精度地分离热反射，保留场景中的物体，并且比现有方法获得更好的性能。