Eddy current pulsed thermography (ECPT) applies induction heating and a thermal camera for non-destructive testing and evaluation (NDTE). Because of the variation in resultant surface heat distribution, the physical mechanism that corresponds to the general behavior of ECPT can be divided into an accumulation of Joule heating via eddy current and heat diffusion. However, throughout the literature, the heating mechanisms of ECPT are not given in detail in the above two thermal phenomena and they are difficult to be separated. Nevertheless, once these two physical parameters are separated, they can be directly used to detect anomalies and predict the variation in material properties such as electrical conductivity, magnetic permeability and microstructure. This paper reports physical interpretation of these two physical phenomena that can be found in different time responses given the ECPT image sequences. Based on the phenomenon and their behaviors, the paper proposes a statistical method based on single channel blind source separation to decompose the two physical phenomena using different stages of eddy current and thermal propagation from the ECPT images. Links between mathematical models and physical models have been discussed and verified. This fundamental understanding of transient eddy current distribution and heating propagation can be applied to the development of feature extraction and pattern recognition for the quantitative analysis of ECPT measurement images and defect characterization.