Ambient Temperature Invariant Infrared Face Recognition Based on Discrete Wavelet Transform

The influence of ambient temperature is a principal challenge of the use of thermal face images for face recognition. This paper proposes a weighted wavelet sub-bands method for infrared face recognition under variable ambient temperatures. The wavelet transform can decompose an infrared image into different levels of approximation and detail sub-bands. The influence of ambient temperature on different sub-bands is different. Based on discrimination criterion, the optimization model is constructed to strengthen the features robust to ambient temperature, and weaken the features vulnerable to ambient temperature. By minimizing the criterion function, we can assign the optimal weights to corresponding sub-bands. Finally, the optimization method is applied to variable ambient temperatures infrared face recognition to verify its efficiency. The experiments demonstrate that the proposed ambient temperature invariant method is feasible and can significantly improve performance of infrared face recognition.