Traditional ghost imaging applies correlated algorithms to reconstruct the image of an object. However, it fundamentally requires some spatial distributions of the correlated light beam, e.g. random illumination, which hardly exists in reality. Here, different from the localized analysis used in the traditional ghost imaging, a spatial and temporal global analysis of the whole measurements is proposed. Therefore, we demonstrate a new ghost imaging modality, called adaptive ghost imaging (AGI), that utilizes the difference of successive frames as the correlation pattern to generate the image. As a result, AGI can work with any varying illuminations including, but not limited to, random illumination. We believe that AGI will make the ghost imaging easier, more applicable and closer to reality.