FTIR spectral imaging enhancement for teacher's facial expressions recognition in the intelligent learning environment

Fourier transform infrared (FTIR) spectral imaging is a valuable tool for rapid identification of the real/fake human face in a single spectral image. However, it often suffers from the problems of band overlap and random noise. In this work, we have developed a spectral resolution enhancement algorithm via the contourlet transforms regularization for FTIR spectral imaging. First, we apply the contourlet transform to the observed FTIR spectrum and the ground-truth one, and compare their distributions of the contourlet coefficients. It shows that the distribution of the ground-truth spectrum is sparser than that of the observed one. Then, an infrared spectral resolution enhancement model is built to regularize the distribution of the observed FTIR spectrum by total variation regularization. Furthermore, an alternating scheme is exploited to estimate the instrument function and the latent FTIR spectrum, and it shows excellent convergence and stability. Experimental results show the effectiveness of the proposed approach in suppressing random noise and preserving spectral details in the FTIR spectral resolution enhancement problems. Thus, the proposed method leads the high-resolution FTIR spectrum as a more efficient tool for the recognition of teacher's facial expressions in the intelligent learning environment.