An improved sparse LS-SVR for estimating illumination

Zhenmin Zhu; Zhaokang Lv; Baifen Liu

doi:10.1117/12.2197082

6 July 2015 An improved sparse LS-SVR for estimating illumination

Zhenmin Zhu, Zhaokang Lv, Baifen Liu

Proceedings Volume 9631, Seventh International Conference on Digital Image Processing (ICDIP 2015); 963107 (2015) https://doi.org/10.1117/12.2197082
Event: Seventh International Conference on Digital Image Processing (ICDIP15), 2015, Los Angeles, United States

Abstract

Support Vector Regression performs well on estimating illumination chromaticity in a scene. Then the concept of Least Squares Support Vector Regression has been put forward as an effective, statistical and learning prediction model. Although it is successful to solve some problems of estimation, it also has obvious defects. Due to a large amount of support vectors which are chosen in the process of training LS-SVR , the calculation become very complex and it lost the sparsity of SVR. In this paper, we get inspiration from WLS-SVM(Weighted Least Squares Support Vector Machines) and a new method for sparse model. A Density Weighted Pruning algorithm is used to improve the sparsity of LS-SVR and named SLS-SVR(Sparse Least Squares Support Vector Regression).The simulation indicates that only need to select 30 percent of support vectors, the prediction can reach to 75 percent of the original one.

Citation Download Citation

Zhenmin Zhu, Zhaokang Lv, and Baifen Liu "An improved sparse LS-SVR for estimating illumination", Proc. SPIE 9631, Seventh International Conference on Digital Image Processing (ICDIP 2015), 963107 (6 July 2015); https://doi.org/10.1117/12.2197082

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available