SSIF: Learning Continuous Image Representation for Spatial-Spectral Super-Resolution

Existing digital sensors capture images at fixed spatial and spectral resolutions (e.g., RGB, multispectral, and hyperspectral images), and each combination requires bespoke machine learning models. Neural Implicit Functions partially overcome the spatial resolution challenge by representing an image in a resolution-independent way. However, they still operate at fixed, pre-defined spectral resolutions. To address this challenge, we propose Spatial-Spectral Implicit Function (SSIF), a neural implicit model that represents an image as a function of both continuous pixel coordinates in the spatial domain and continuous wavelengths in the spectral domain. We empirically demonstrate the effectiveness of SSIF on two challenging spatio-spectral super-resolution benchmarks. We observe that SSIF consistently outperforms state-of-the-art baselines even when the baselines are allowed to train separate models at each spectral resolution. We show that SSIF generalizes well to both unseen spatial resolutions and spectral resolutions. Moreover, SSIF can generate high-resolution images that improve the performance of downstream tasks (e.g., land use classification) by 1.7%-7%. While the physical world is continuous, most digital sensors (e.g., cell phone cameras, multispectral or hyperspectral sensors in satellites) can only capture a discrete representation of continuous signals in both spatial and spectral domains (i.e., with a fixed number of spectral bands, such as red, green, and blue). In fact, due to the limited energy of incident photons, fundamental limitations in achievable signal-to-noise ratios (SNR), and time constraints, there is always a trade-off between spatial and spectral resolution (Mei et al., 2020; Ma et al., 2021) However, ML models are typically bespoke to certain resolutions, and models typically do not generalize to spatial or spectral resolutions they have not been trained on.