Machine Learning
On the effects of dimensionality reduction on high dimensional similarity search
PODS '01 Proceedings of the twentieth ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems
Unsupervised Learning Using Locally Linear Embedding: Experiments with Face Pose Analysis
ICPR '02 Proceedings of the 16 th International Conference on Pattern Recognition (ICPR'02) Volume 1 - Volume 1
Think globally, fit locally: unsupervised learning of low dimensional manifolds
The Journal of Machine Learning Research
Hi-index | 0.00 |
To overcome the problems associated with high dimensionality, such as high storage and classification time, dimension reduction is usually applied to the vectors to concentrate relevant information in a low dimension. Locally Linear Embedding (LLE) is a well-known dimension reduction scheme. However, it works with vectorized representations of images and does not take into account the spatial locality relation information of images, thus some information will be lost. In this paper, a new dimension reduction scheme, called Small Matrix Vector Locally Linear Embedding (SMVLLE), is presented. Using SMVLLE which is based on small matrix cover for dimension reduction can reduce the loss of spatial locality relation information among image pixels, because this scheme works directly with images in their native state. Experiments on handwritten digit images and texture images show that SMVLLE is superior to LLE in terms of quality of the dimension reduction.