Impact of training data on LMMSE demosaicing for Colour-Polarization Filter Array

Abstract

Linear minimum mean square error can be used to demosaic images from a colour-polarization filter array sensor. However, the role of training data on its performance is yet an open question. We study the model selection using crossvalidation techniques. The results show that the training model converges quickly, and that there is no significant difference in training the model with more than 12 images of approximately 1.5 megapixels. We also found that the selected trained model performs better compared to a dedicated Colour-Polarization Filter Array demosaicing algorithm in terms of Peak Signal-to-Noise Ratio.