面向缺失像素图像集的修正拉普拉斯特征映射算法

Modified Laplacian Eigenmap Algorithm for Missing Pixels Image Set

针对缺失像素图像集,提出修正的拉普拉斯特征映射算法.该算法将缺失像素图像集看成向量集,利用向量之间的余弦相似度衡量缺失像素图像之间的距离,提出一种新的权值构造函数,并在多组标准测试数据集上进行实验.结果表明:修正的拉普拉斯特征映射算法可以很好地挖掘缺失像素图像数据集的内在流形结构,减弱缺失像素带来的不良影响.

In this paper,we propose a modified laplacian eigenmaps algorithm for the missing pixel images.The algorithm takes the missing pixel image set as a vector set,then using the cosine similarity between vectors to measure the distance between missing pixel images.Further,a new weight constructor function is proposed.Experiments on several sets of standard test data sets show that the modified laplacian eigenmaps algorithm can well excavate the intrinsic manifold structure of the missing pixel images and weaken the negative effects of missing pixels.