The low-resolution images produced by infrared systems can make it difficult for night-vision users to estimate distances or identify far objects. “Resolution enhancement” attempts to address these problems by converting several low-quality images into a single, higher-quality version. Unfortunately, conventional methods of reconstructing these low-quality images are computationally intense, making them too expensive or too slow for real-time, affordable use.
Researchers in ECE’s Computer Vision Laboratory have developed a new resolution enhancement technique based on super-resolution reconstruction (SRR). The new method is fast and memory efficient, providing performance comparable to previous techniques but with dramatically reduced computation requirements, according to Ph.D. student Jae Cha and his advisor, Lynn Abbott.
Their closed-form, one-pass method is much faster than existing iterative approaches. In several experiments, they demonstrated a 10-fold speed-up in computation time. The speed, combined with low memory requirements, make this method suitable for use in a low-cost distance-estimation system for infrared applications.