关键词: 数据降维, 流形学习, 局部线性嵌入, 图像检索

Abstract: Locally Linear Embedding（LLE） algorithm is one of the non-linear dimensionality reduction methods which are based on manifold learning.In LLE，each sample point is reconstructed from a linear combination of its nearest neighbors.However，different number of neighbors will produce different reconstruction errors，which will make the result different directly.This paper structures the approximate reconstruction coefficient making use of their category information which is obtained by clustering，and proposes an improved algorithm.The proposed algorithm can reduce the influence of the number of neighbors efficiently and the probability of the database is retained.This is confirmed by experiments on both synthetic and real-world data.

Key words: data dimensionality reduction, manifold learning, locally linear embedding, image retrieval