Recognizing graphic objects from binary images is an important task in many real-life applications. Generally, there are two ways to do the graphics recognition: one-step methods and two-step methods. The former recognizes graphic objects from binary images directly, while the latter consists of vectorization and postprocessing. Neither of them is perfect enough to handle all difficulties. This paper first reviews popular graphics recognition methods to understand their advantages and disadvantages. Next, the performance comparison between two classes of methods is made in two important aspects: the time efficiency and the graphics quality, and the experimental results of time-efficiency comparison of 7 popular methods are also reported. Finally, we propose a new hybrid graphics-recognition paradigm to integrate the advantages of both one-step methods and two-step methods and minimize their disadvantages. The proposed paradigm is capable of recognizing straight lines, arcs, circles and curves efficiently, and is helpful for extracting text images in text-graphics touching cases.