Context-aware computing is getting emergent in recent years. It focuses on how the system can deliver suitable services as the change of context information. The services rely on the available context information. The development of new context information is an important task for extending the application domain of context-aware computing. In this paper, an image-based approach for generating the direction information of a mobile device is proposed. The direction information is generated by matching the real-time image from a mobile device and 360? surrounding image at a specific site. The matching is based on the skylines of two images. The skyline is detected and encoded into a series of codes (SCode). A set of turning points (TPs) is then selected from the SCodes. A dynamic threshold technique is also designed to overcome the influence of cloud, brightness, etc. From the experimental results of six different sites, the average accuracy is 84 percent when the inaccuracy level is ten pixels. The average execution time is only 1.6 second on Pocket PC with 400MHz CPU. It illustrates that the proposed approach is effective and efficient to generate direction information for context-aware computing.