Rethinking Compressive Sensing: A Compression Framework for Video Super-Resolution

Ruthy Katz, Adi Teitel, Moran Mordechay, Adi Falik, Eli Bery, Maya Mayberg; Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR) Workshops, 2025, pp. 739-748

Sensors in mobile phones have significantly increased in dimensions, reaching an impressive 200M pixels. However, the output resolution of images and videos stored on a device is still constrained. In most modern high-resolution imaging sensors, pixel binning is performed to maintain a good signal-to-noise ratio (SNR) and accommodate memory and bandwidth resource limitations. This presents a challenge, as extensive spatial compression leads to irreversible data loss. Although Video Super-Resolution (VSR) methods can be applied to reconstruct high-resolution videos, they may still lead to unsatisfactory reconstruction results. To address this challenge, we rethink the role of Compressive Sensing (CS) in the video domain. We propose a novel on-sensor sampling scheme and examine different dynamic binning mask layouts with temporal variations. For reconstruction, we use conventional VSR models and demonstrate that combining them with our compression strategy enhances image quality. Extensive experiments on benchmark datasets confirm that our method improves reconstruction quality both quantitatively and qualitatively.