A correlation graph approach for unsupervised manifold learning in image retrieval tasks

摘要: Effectively measuring the similarity among images is a challenging problem in image retrieval tasks due to difficulty of considering dataset manifold. This paper presents an unsupervised manifold learning algorithm that takes into account intrinsic geometry for defining more effective distance images. The structure modeled terms Correlation Graph (CG) and analyzed using Strongly Connected Components (SCCs). While adjacency provides precise but strict relationship, analysis expands these relationships geometry. A large rigorous experimental evaluation protocol was conducted different tasks. experiments were datasets involving various descriptors. Results demonstrate can significantly improve effectiveness systems. presented approach yields better results than methods recently proposed literature.