Study Of Constrained Clustering In Face Annotation System

With the development of technology, the requirement of face annotation system is increasing day by day. A feature of face annotation system is face-based clustering. Unsupervised clustering algorithms have a limited effect on this problem, so constraint clustering algorithms using constraints are presented.Constraints are usually in the form of pairwise relationships as must-link and cannot-link. The introduction of such constraints in a clustering algorithm can enhance the clustering effect to a certain extent. In this paper, we study the con-straint clustering in face annotation system. Based on the work of constrained clustering and face pattern recognition, we make full investigation and experi-ment, and then propose a constrained construction method, and design the cor-responding face annotation system, find the most suitable algorithm through simulation experiments, and complete the demo system program. This system can obtain a large number of high-quality constraints by user simple annotation,and then enhance the effect of constrained clustering. The system uses unsuper-vised clustering when there are no constraints, and constrained clustering after constraints. For the clustering results, the user only needs to mark the first error in the largest cluster, and can construct a large number of constraints, which can guide the constrained clustering well. This system can reduce the workload of user, facilitate face annotation and image management.