A multi-modal and multi-scale non-local means (M3S-NLM) method is proposed to extract atomically resolved spectroscopic maps from low signal-to-noise (SNR) datasets recorded with a transmission electron microscope. This method improves upon previously tested denoising techniques as it takes into account the correlation between the dark-field signal recorded simultaneously with the spectroscopic dataset without compromising on the spatial resolution. The M3S-NLM method was applied to electron energy dispersive X-ray and electron-energy-loss spectroscopy (EELS) datasets. We illustrate the retrieval of the atomic scale diffusion process in an Al1-xInxN alloy grown on GaN and the surface oxidation state of perovskite nanocatalysts. The improved SNR of the EELS dataset also allows the retrieval of atomically resolved oxidation maps considering the fine structure absorption edge of LaMnO3 nanoparticles.