The real-time and daytime color sky images captured by ground-based cloud imager are object of study，the time series images are processed and analyzed by digital image processing technologies.The estimation of cloud future motion is realized by using cloud extraction algorithm and cloud tracking and learning algorithm.The obstruction state of sun in the future time is predicted combining the location of sun in the cloud images in a day，and then the change of irradiance and photovoltaic power are predicted.The results show that the very short-term prediction model based on the ground-based images is feasible and practical，which provides a significant way to forecast for the photovoltaic power output within0-4 hours.