Intelligent Database Systems Lab 國國國國國國國國 National Yunlin University of Science and Technology 1 Fast-Learning Adaptive- Subspace Self-Organizing Map: An Application to Saliency- Based Invariant Image Feature Construction Presenter : You Lin Chen Authors : Huicheng Zheng, Member, IEEE, Grégoire Lefebvre, and Christophe Laurent 2007.WI.7
16
Embed
Presenter : You Lin Chen Authors : Huicheng Zheng, Member, IEEE, Grégoire Lefebvre,
Fast-Learning Adaptive-Subspace Self-Organizing Map: An Application to Saliency-Based Invariant Image Feature Construction. Presenter : You Lin Chen Authors : Huicheng Zheng, Member, IEEE, Grégoire Lefebvre, and Christophe Laurent. 2007.WI.7. Outline. Motivation - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Intelligent Database Systems Lab
國立雲林科技大學National Yunlin University of Science and Technology
The traditional learning procedure of the ASSOM involves computations related to a rotation operator matrix.
The rotation computations which not only is memory demanding, but also has computational load quadratic to the input dimension.
3
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
4
2 4.
83.3
22+4.82+3.32
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Objectives
In this paper will show that in the ASSOM learning which leads to a computational load linear to both the input dimension and the subspace dimension.
we are also interested in applying ASSOM to saliency-based invariant feature construction for image classification.
5
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Methodology_1
6
Kohonen’s ASSOM learning algorithm
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Methodology_1
7
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Methodology_1
8
Robbins–Monro stochastic
approximation
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Methodology_1
9
BFL-ASSOM
FL-ASSOMBFL-ASSOM
ASSOM
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
10
Experiments_1
The input episodes are generated by filtering a white noise image with a second-order Butterworth filter. The cutoff frequency isset to 0.6 times the Nyquist frequency of the sampling lattice.
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments_1
11
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Methodology_2
12
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Methodology_2
13
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments_2
14
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
15
Conclusion
The ASSOMis useful for dimension reduction, invariant feature generation, and visualization.
BFL-ASSOM, where the increment of each basis vector is a linear combination of the component vectors in the input episode.
The SPMAS showed promising performance on a ten-category image classification problem