In reverse engineering, in order to get the entire point data of the object surface, which must be merged from point data cloud obtained from different angles. And in order to acquire exactly merge effect, a method to stick reference points on object surface to acquire base point. Thus, the precision of center extraction of reference point affect the merge result. Firstly a new method to center extraction of reference point is presented based on the characteristics to the images in reverse engineering in this paper. And the new method is that the region recognition algorithm combining to least squares algorithm. Secondly, in order to improve the computational efficiency, the modified sequential algorithm is applied to optimize the region recognition. Lastly, the availability and accuracy to the new method are proved by the experiment.