In multiple object identification method,
objects in the query image are detected and identified by clustering the keypoint matches into individual objects.
For each keypoint match, an object center position is estimated using scale and orientation associated with the keypoint.
The estimated object centers are clustered using mean shift, and each cluster is regarded as a match to the object in the database image.
Given n estimated object center positions and n initial positions of clusters , mean shift clusters estimated object centers by updating using a difference vector of and a center of gravity of near .
The computational complexity of mean shift increases with the square of .
Therefore, it takes an enormous amount of time for identifying a large amount of identical objects, such as shown in Fig. 1.