The Application Of Manifold Learning Algorithms In Pattern Recognition

Dimensionality reduction plays an important role in the field of computer vision,machine learning, and pattern recognition. Dimensionality reduction method hasachieved good development and progress, especially in recent years raised manifoldlearning methods. We just talk about neighborhood graph construction and incrementallearning study in manifold learning, mainly to dynamically determine the size of theneighborhood, estimating neighborhood using local geodesic distance, incrementalalgorithm processing new data.1. We present a modified ISOMAP algorithm to adaptively determine the size of theneighborhood, which overcomes the sensitivity to the initial size of the neighborhood,andcan deal with a large number of non-uniform distribution manifold and has betterrobustness of image noise and geometric. And we also discuss another ISOMAPalgorithm to estimate neighborhood using local geodesic distance. When building theneighborhood graph of each of the input data points, this algorithm defines a newmeasure and uses it to optimize the neighborhood of each of the input data points. Andto some extent it overcomes the problems caused by short-circuit and it also has betterrobustness of image noise and geometric.2. We propose an improved incremental ISOMAP algorithm to simultaneouslysolve the two problems of the classic ISOMAP algorithms. When building theneighborhood graph of each of the input data points, the algorithm defines a newmeasure and uses it to optimize the neighborhood of each of the input data points. Andit also uses incremental learning method instead of the batch learning. So it overcomesthe disadvantage that we must get all of the data samples before using them and whenthe new data is obtained, we must recalculate all of the data. In the face recognition test,it achieves a higher recognition rate and saves a lot of computing time.