Colorization methods using deep neural networks have become a recent trend. However, most of them do not allow user inputs, or only allow limited user inputs (only global inputs or only local inputs), to control the output colorful images. The possible reason is that it's difficult to differentiate the influence of different kind of user inputs in network training. To solve this problem, we present a novel deep colorization method, which allows simultaneous global and local inputs to better control the output colorized images. The key step is to design an appropriate loss function that can differentiate the influence of input data, global inputs and local inputs. With this design, our method accepts no inputs, or global inputs, or local inputs, or both global and local inputs, which is not supported in previous deep colorization methods. In addition, we propose a global color theme recommendation system to help users determine global inputs. Experimental results shows that our methods can better control the colorized images and generate state-of-art results.