Seismology is the study of the vibration of the Earth. Seismologists pay much attention to the main source of Earth vibration: earthquakes. But also other seismic sources, like mining blasts, ocean storms and windmills, are studied. All these sources induce seismic waves, which can eventually be recorded as ground vibrations. These seismic records contain not only information about the sources, but also about the part of the Earth through which the waves have propagated. This thesis focuses on a main subclass of seismology: seismic imaging. With seismic imaging, seismic records are studied with the aim to unravel the structure and composition of the Earth. Seismic imaging finds its main use in extracting information about the solid Earth that is within our reach. With the current state of the art, we can only mine minerals from the upper 0.2 percent. Imaging the deep (i.e., unreachable) Earth finds its main use in hazard assessment. Through imaging the deep Earth, we can better understand the dynamics of our planet, manifested in, amongst others, earthquakes and volcanism. By forecasting how the Earth will reshape, we can sensibly adapt. Seismic characterization of the subsurface also improves our ability to assess the direct impact of earthquakes. A large part of the current deep-Earth images are obtained through tomographic inversion. These images are successful for identifying large-scale structural anomalies, like (remnants of) subducting slabs. However, they lack the resolution to accurately image discontinuities in space. Sharp discontinuities, like the interface between sedimentary and crystalline rock, lead to conversions and reflections of seismic waves. It is those reflections that are most suitable for imaging the discontinuities. The technique developed to do this is called seismic reflection imaging (SRI). Though SRI is considered to be an advanced technique for imaging the subsurface, the dense and wide source (and receiver) distribution requirement thus far limited the use for deep-Earth imaging. Deep-Earth imaging is done primarily with natural sources. Their limited and irregular distribution and their unpredictable occurrence time is inconvenient for SRI. In this thesis we work on a methodology to alleviate the stringent requirement on the source distribution. This methodology is called seismic interferometry (SI). SI is a remapping operation. With SI, the responses from many sources are combined to create the response as if there were a source at a receiver position. The response from this "virtual" source could contain both surface waves and body waves. We apply SI such that especially (reflected) body waves are retrieved, because it are these waves that are used for SRI. When we apply SI to a regularly-spaced array of receivers, we could turn uncontrolled natural sources into a well-organized succession of virtual sources, which is amenable for SRI. The aim of this work is to improve the imaging of the interior of the Earth by the application of body-wave SI to naturally induced seismic data. A prior study focused on the exploration-scale (i.e., the part of the Earth that is being mined). In this thesis, we focus on low-frequency waves (< 2 Hz), which contain especially information about the deep Earth. In Part I we consider applications to earthquake recordings. In Part II we evaluate the retrieval of reflections using microseisms. Part I starts by looking into the global-scale configuration. For this configuration, two seismic stations (or two arrays of stations) may be located anywhere on the globe and reflectors may be located anywhere in depth. We derive and numerically test relations to retrieve the complete response between two stations, using worldwide seismicity. The SI relations are valid for a closed entity for seismic waves, like the Earth is by approximation. The relations do not account for inelastic losses, by which eventually all seismic energy is transformed to heat. Though normally of negligible influence, the SI relations do account for the rotation of the Earth. In practice, the distribution of the larger earthquakes is not wide enough to retrieve complete responses between any two points on the Earth. However, still relevant primary reflections can be retrieved with a limited dense distribution of seismicity, like from the Tohoku (2011) aftershock sequence. Part I proceeds by zooming in to the lithospheric-scale configuration. The lithosphere is the part of the solid Earth where most (seismic) activity takes place. It makes up the crust and the upper part of the mantle. It is this part of the Earth that, broken into plates and fueled by convection in the mantle, undergoes birth (at midoceanic ridges), collisions (resulting in stunning topography) and death (subduction back into the mantle). Inspecting the Earth in depth unveils a part of the history of the lithosphere and gives an indication of the processes still ongoing. Information about the structure of the lithosphere is hidden in the coda of wavefields that arrive due to distant seismicity. We test a few different SI approaches to extract this information. We use synthetic data with similar characteristics as field data recorded during the Laramie broad-band experiment (2000-2001). This was an array of seismic stations in Wyoming, USA, to study an suture zone. We evaluate the requirements for obtaining multi-offset reflection responses. These multi-offset reflections are important for obtaining a velocity model. We estimate this model for the subsurface below the Laramie array and we use it to map the extracted reflections to a reflectivity image of the lithosphere. In the remainder of part I, we focus again on the lithospheric scale. This time, we assume that a velocity model is already available through other means. Consequently, only single-offset reflection responses are required to make an image. We use global phases to obtain zero-offset reflection responses. We show the robustness of the method with data from the Hi-CLIMB experiment (2002-2005). This was a large and well-sampled seismic array, passing the Himalayas and a significant part of the Tibetan Plateau. The successive application of SRI leads to an image of the Indian-Eurasian Plate collision. The rediscovery of body waves in low-frequency noise (