The surface temperature retrieval of high-temperature objects shielded by strong diffusive mist plays a critical role in aerospace, metallurgical and many other industries. Traditional radiation temperature measurements under extreme conditions usually cause substantial errors because of the strong extinction and scattering by mist during light propagation. The current infrared temperature retrieval methods often use single-channel or double-channel non-imaging temperature measurement strategies. In such methods, the temperature is corrected using either pre-measured or real-time mist parameters, and the results are analyzed and evaluated according to radiative transfer theory. Based on the calculated spectral radiation characteristics of mist, this paper proposed an infrared imaging temperature measurement method. A temperature retrieval model is also built based on radiative transfer theory while the adjacent effect is considered. The exact surface temperature distribution can be retrieved while the parameters of the diffusive mist medium remain unknown. During a typical temperature retrieval process, the radiative temperature distribution is firstly calculated according to the infrared images in three different channels calibrated by the pre-acquired calibration and emissivity data. Then the exact temperature is retrieved according to this non-linear temperature retrieval model. A three-channel infrared temperature retrieval system is designed, with its three channels centered at 8.8, 10.7, and 12.0 μm, respectively. Three identical long-wave infrared focal plane detector is applied, which can simultaneously photograph the high-temperature object. Besides, an experimental verification device is assembled to test the performance of the three-channel infrared system based on a high-temperature blackbody and a home-made mist generator. The results prove that long-wave infrared shows a higher interference resistance capacity than mid-wave infrared. This three-channel device and the temperature retrieval model reduce the image distortion caused by mist and show an average temperature retrieval error of ca. 7% at 1 000, 1 100, and 1 200 ℃ conditions. This method is suitable for both high-temperature blackbody and graybody, while the pre-acquisition of the mist parameters is not required. The temperature retrieval method based on multi-spectral infrared imaging proposed in this article shows universal applicability and considerable innovativeness.