Respiratory motion is a major challenge in cardiac magnetic resonance imaging (MRI) and contemporary state-of-the-art motion compensation strategies like diaphragmatic navigators still suffer from sub-optimal time efficiency. In response, k-space- based one-dimensional self-navigation techniques have recently been developed that extract respiratory-induced motion of the heart directly from the imaging data themselves for subsequent motion correction on a beat-to-beat basis [1]. This affords the advantage of 100% scan efficiency while meticulous plan scanning and navigator placement can be avoided. In the present study, this concept was advanced to the next level by implementing an image- based self-navigation technique that incorporates compressed sensing and allowing for multi- dimensional motion correction. The new approach was investigated using computer simulations of a moving heart phantom before it was implemented on a 3T human scanner. In 12 healthy adult human subjects, the performance of this methodology was then quantitatively ascertained in comparison to free-breathing coronary MRI, both with and without conventional respiratory navigators.