To get stable biological features from the time-elapse infrared face, a new construction method of blood perfusion is proposed based on bio-heat transfer, which can be applied to face recognition. Firstly, according to the classic bio-heat transfer equation (the Pennes Equation), the blood perfusion rate in different positions can be computed based on the new blood perfusion model (PBPM). Then, due to the low-resolution of the infrared images, the feature extraction method (Discrete Cosine Transformation) is chosen to get the principle features in the blood perfusion image. The experiment results demonstrate that compared traditional methods, the blood perfusion model proposed in this paper is more stable, the infrared face recognition method based on the new blood perfusion model is robust to both same-session data and time-elapse data.