In the diagnosis of Printed Circuit Board (PCB) based on infrared radiation, emissivity correction must be performed on the thermal image to retrieve the true temperature distribution on the PCB surface. Taking the measured PCB surface temperature as a combination of its true temperature distribution and its emissivity distribution, a non-linear filtering method was proposed to separate the true temperature from this combination based on the differences in the continuity of these two contributions. As a key step of the proposed method, the wavelet implementation of discontinuities detection in thermal image was also discussed. Experimental results indicated that the true temperature distribution can be estimated precisely by the proposed method. It is also shown that the proposed method can be used in other applications such as heat source recognition on PCB infrared thermal image and emissivity estimation of unknown materials.