Combustion efficiency of industrial furnaces is effectively regulated based on the oxygen concentration in the flue gases. According to the concept of control, the more energy conservation can be reached by reducing the control variance due to the measurement delay. The image-based predictive control is proposed in this paper. A flame identification technology using fuzzy logics is applied and calculation of flame features is processed simultaneously. A control strategy combining the prediction model of oxygen concentration and use of an ideal PI controller has put into practice for the optimization of combustion operation. This paper compares and assesses the performance of analyzer-based and image-based control strategies as well as the application to an experimental furnace. The results show the image-based predictive control has higher efficiency than the traditional control strategy.