Various speckle-based computational imaging techniques that exploit the ability of scattering media to transfer hidden information into the speckle pattern have recently been demonstrated. Current implementations suffer from several drawbacks associated with the use of conventional scattering media (CSM), such as their time-consuming characterization, instability with time, and limited memory-effect range. Here we show that by using a random dielectric metasurface diffuser (MD) with known scattering properties, many of these issues can be addressed. We experimentally demonstrate an imaging system with the ability to retrieve complex field values using a MD and the speckle-correlation scattering matrix method. We explore the mathematical properties of the MD transmission matrix such as its correlation and singular value spectrum to expand the understanding about both MDs and the speckle-correlation scattering matrix approach. In addition to a large noise tolerance, reliable reproducibility, and robustness against misalignments, using the MD allows us to substitute the laborious experimental characterization procedure of the CSM with a simple simulation process. Moreover, dielectric MDs with identical scattering properties can easily be mass-produced, thus enabling real-world applications. Representing a bridge between metasurface optics and speckle-based computational imaging, this work paves the way to extending the potentials of diverse speckle-based computational imaging methods for various applications such as biomedical imaging, holography, and optical encryption.