Image Recovery in the Infrared Domain via Path-Augmented Compressive Sampling Matching Pursuit

Tegan H. Emerson, Colin C. Olson, Anthony Lutz; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 0-0

We consider compressive sensing as a means of acquiring high-resolution images from low-cost, low-resolution sensors in the infrared domain. In particular, we reduce errors arising from basis mismatch between the observed image and the signal model by modifying a baseline matching pursuit recovery algorithm. Specifically, we introduce a modification to the analysis step which seeks to find more representative image atoms by searching over a 2-Wasserstein geodesic formed between the two most-correlated atoms at that step. We test our extension by quantifying recovery performance on an ensemble of representative infrared maritime scenes and find improvement over baseline when measured using PSNR, SSIM, and a metric that quantifies global edge recovery performance. We find that the most notable gains occur for very low sparsity levels which favors reduced computational load for the recovery.