Spectral Estimation Of Image Features Manifold Learning Method

Manifold learning methods have gradually become an important branch in the field of machine learning. Aiming at questions in image feature manifold, this thesis puts forward spectrum estimation learning frame of image feature manifold which includes:(1) analyzing relations between relevant spectrum theories on topological invariance of image manifold and manifold learning, and giving solid theoretical proofs to these relations;(2) establishing manifold dimensionality reduction algorithm with geodesic distance as proximity measurement and providing confirmation of specific examples;(3) putting forward spectrum estimation learning algorithm of topological invariance of image feature manifold, and applying this algorithm to dimensionality deduction, clustering, face recognition and other fields, and proving validity of the algorithm herein through examples.In conclusion, starting from image feature manifold, this thesis has put forward relevant learning algorithm and has made some achievements. However, there are still many question demanding further researches, such as selection of feature vector and online learning on samples of unknown clustering numbers.