In ground-penetrating radar imaging, the classic back-projection (BP) algorithm has an excellent reputation for imaging in layered media with convenience and robustness. However, it is time-consuming and generates many artifacts, which have adverse effects on detection and recognition. A self-correlation back-projection (SBP) algorithm is proposed, which is fast imaging and can distinguish the object’s shape. It improves the existing BP algorithms in the following aspects. First, the reflection echo signals of a specific imaging point obtained from its nearest exploration point have high correlation with the one from its multiple nearest neighbors. By setting up a correlation threshold, the valid echo information sequence of the imaging points can be adaptively chosen, which enables the SBP algorithm to have faster calculation speed and better resolution. Then, the imaging result is amended by using a depth energy compensation algorithm. It can improve the imaging resolution of the deep underground objects. The experimental results show that the proposed SBP algorithm is superior to the existing BP algorithms in terms of computing speed and imaging accuracy, which can effectively recover objects with complex shapes. It has a significant advantage in providing a rough outline of buried objects without prior knowledge of the velocity distribution.