The time-of-flight (TOF) camera has recently received significant attention due to its small size, low cost, and low-power consumption, which can be widely used in fields such as automatic navigation and machine vision. The TOF camera can calculate 3D information of targets with dozens of frames per second. However, poor accuracy still exists in the presence of various inevitable disturbances. In particular, the imaging distance and object reflectivity are remarkable factors. In this study, the depth imaging conditions, including ambient light, detection distance, and object reflectivity, are theoretically analyzed using differential entropy. Because many coupled factors disturb the imaging accuracy simultaneously, we propose a type of supervised learning machine, entropy-based k-nearest neighbor, based on differential entropy. Experiments show that this method can significantly improve the accuracy of depth data obtained by a TOF camera.