Metasurface-based computational imaging has gained significant attention due to its compactness, multifunctionality, and subwavelength coding capabilities. By integrating computational imaging techniques, researchers have explored the extended capabilities of metasurfaces, enabling a wide range of imaging methods. This review summarizes recent progress in metasurface-based imaging techniques from the perspective of computational imaging, categorizing them into three main groups: (i) conventional metasurface design using canonical methods, (ii) computation introduced independently in the imaging process or postprocessing, and (iii) end-to-end computation-optimized imaging systems based on metasurfaces. The review highlights the advantages and challenges of each technique and discusses their potential and future prospects. Metasurfaces, composed of subwavelength nanostructures arranged in specific patterns, can interact with light and electromagnetic waves in a highly controlled manner. They enable diverse imaging techniques, including microscopic imaging, hyperspectral imaging, full-Stokes polarization imaging, and full-space 3D imaging. However, challenges such as nonidealities introduced by metasurfaces and the need for a general, scalable imaging framework remain. Computational imaging, at the intersection of optics, electronics, signal processing, and machine learning, aims to incorporate computational techniques into imaging and reconstruction processes. It relaxes optical system design constraints and expands imaging capabilities. Computational strategies have been integrated into metasurface-based imaging, significantly enhancing current imaging systems. The review discusses metasurface-based computational imaging in terms of plenoptic dimension modulation, including spectrum, polarization, phase, and compound modulation. It categorizes existing metasurface-based computational imaging frameworks based on where computation is introduced, such as computational illumination, sensing, and reconstruction. The review also addresses the challenges in building metasurface-based computational imaging systems and discusses future research directions. The review highlights the potential of metasurfaces in spectral imaging, polarization modulation, and depth/angle modulation. It discusses various techniques for achieving achromatic imaging, hyperspectral imaging, full-Stokes polarization imaging, and wide-angle imaging. The review also explores compound modulation, which involves coordinated manipulation of various light field properties to achieve specific imaging goals. Overall, the integration of computational imaging with metasurfaces has enabled significant advancements in imaging technology, offering compact, flexible, and high-performance imaging systems. The review provides insights into the current state of metasurface-based computational imaging and identifies key challenges and future research directions.Metasurface-based computational imaging has gained significant attention due to its compactness, multifunctionality, and subwavelength coding capabilities. By integrating computational imaging techniques, researchers have explored the extended capabilities of metasurfaces, enabling a wide range of imaging methods. This review summarizes recent progress in metasurface-based imaging techniques from the perspective of computational imaging, categorizing them into three main groups: (i) conventional metasurface design using canonical methods, (ii) computation introduced independently in the imaging process or postprocessing, and (iii) end-to-end computation-optimized imaging systems based on metasurfaces. The review highlights the advantages and challenges of each technique and discusses their potential and future prospects. Metasurfaces, composed of subwavelength nanostructures arranged in specific patterns, can interact with light and electromagnetic waves in a highly controlled manner. They enable diverse imaging techniques, including microscopic imaging, hyperspectral imaging, full-Stokes polarization imaging, and full-space 3D imaging. However, challenges such as nonidealities introduced by metasurfaces and the need for a general, scalable imaging framework remain. Computational imaging, at the intersection of optics, electronics, signal processing, and machine learning, aims to incorporate computational techniques into imaging and reconstruction processes. It relaxes optical system design constraints and expands imaging capabilities. Computational strategies have been integrated into metasurface-based imaging, significantly enhancing current imaging systems. The review discusses metasurface-based computational imaging in terms of plenoptic dimension modulation, including spectrum, polarization, phase, and compound modulation. It categorizes existing metasurface-based computational imaging frameworks based on where computation is introduced, such as computational illumination, sensing, and reconstruction. The review also addresses the challenges in building metasurface-based computational imaging systems and discusses future research directions. The review highlights the potential of metasurfaces in spectral imaging, polarization modulation, and depth/angle modulation. It discusses various techniques for achieving achromatic imaging, hyperspectral imaging, full-Stokes polarization imaging, and wide-angle imaging. The review also explores compound modulation, which involves coordinated manipulation of various light field properties to achieve specific imaging goals. Overall, the integration of computational imaging with metasurfaces has enabled significant advancements in imaging technology, offering compact, flexible, and high-performance imaging systems. The review provides insights into the current state of metasurface-based computational imaging and identifies key challenges and future research directions.