Conditionally positive definite kernels for SVM based image recognition

Sabri Boughorbel, Jean Philippe Tarel, Nozha Boujemaa

Research output: Chapter in Book/Report/Conference proceedingConference contribution

39 Citations (Scopus)

Abstract

Kernel based methods such as Support Vector Machine (SVM) have provided successful tools for solving many recognition problems. One of the reason of this success is the use of kernels. Positive defi niteness has to be checked for kernels to be suitable for most of these methods. For instance for SVM, the use of a positive defi nitekernel insures that the optimized problem is convex and thus the obtained solution is unique. Alternative class of kernels called conditionally positive defi nitehave been studied for a long time from the theoretical point of view and have drawn attention from the community only in the last decade. We propose a new kernel, named log kernel, which seems particularly interesting for images. Moreover, we prove that this new kernel is a conditionally positive defi nitekernel as well as the power kernel. Finally, we show from experimentations that using conditionally positive defi nite kernels allows us to outperform classical positive defi nitekernels.

Original languageEnglish
Title of host publicationIEEE International Conference on Multimedia and Expo, ICME 2005
Pages113-116
Number of pages4
Volume2005
DOIs
Publication statusPublished - 2005
Externally publishedYes
EventIEEE International Conference on Multimedia and Expo, ICME 2005 - Amsterdam, Netherlands
Duration: 6 Jul 20058 Jul 2005

Other

OtherIEEE International Conference on Multimedia and Expo, ICME 2005
CountryNetherlands
CityAmsterdam
Period6/7/058/7/05

Fingerprint

Image recognition
Support vector machines

ASJC Scopus subject areas

  • Engineering(all)

Cite this

Boughorbel, S., Tarel, J. P., & Boujemaa, N. (2005). Conditionally positive definite kernels for SVM based image recognition. In IEEE International Conference on Multimedia and Expo, ICME 2005 (Vol. 2005, pp. 113-116). [1521373] https://doi.org/10.1109/ICME.2005.1521373

Conditionally positive definite kernels for SVM based image recognition. / Boughorbel, Sabri; Tarel, Jean Philippe; Boujemaa, Nozha.

IEEE International Conference on Multimedia and Expo, ICME 2005. Vol. 2005 2005. p. 113-116 1521373.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Boughorbel, S, Tarel, JP & Boujemaa, N 2005, Conditionally positive definite kernels for SVM based image recognition. in IEEE International Conference on Multimedia and Expo, ICME 2005. vol. 2005, 1521373, pp. 113-116, IEEE International Conference on Multimedia and Expo, ICME 2005, Amsterdam, Netherlands, 6/7/05. https://doi.org/10.1109/ICME.2005.1521373
Boughorbel S, Tarel JP, Boujemaa N. Conditionally positive definite kernels for SVM based image recognition. In IEEE International Conference on Multimedia and Expo, ICME 2005. Vol. 2005. 2005. p. 113-116. 1521373 https://doi.org/10.1109/ICME.2005.1521373
Boughorbel, Sabri ; Tarel, Jean Philippe ; Boujemaa, Nozha. / Conditionally positive definite kernels for SVM based image recognition. IEEE International Conference on Multimedia and Expo, ICME 2005. Vol. 2005 2005. pp. 113-116
@inproceedings{902e2b7a82b64ead8603b19cb7a02d07,
title = "Conditionally positive definite kernels for SVM based image recognition",
abstract = "Kernel based methods such as Support Vector Machine (SVM) have provided successful tools for solving many recognition problems. One of the reason of this success is the use of kernels. Positive defi niteness has to be checked for kernels to be suitable for most of these methods. For instance for SVM, the use of a positive defi nitekernel insures that the optimized problem is convex and thus the obtained solution is unique. Alternative class of kernels called conditionally positive defi nitehave been studied for a long time from the theoretical point of view and have drawn attention from the community only in the last decade. We propose a new kernel, named log kernel, which seems particularly interesting for images. Moreover, we prove that this new kernel is a conditionally positive defi nitekernel as well as the power kernel. Finally, we show from experimentations that using conditionally positive defi nite kernels allows us to outperform classical positive defi nitekernels.",
author = "Sabri Boughorbel and Tarel, {Jean Philippe} and Nozha Boujemaa",
year = "2005",
doi = "10.1109/ICME.2005.1521373",
language = "English",
isbn = "0780393325",
volume = "2005",
pages = "113--116",
booktitle = "IEEE International Conference on Multimedia and Expo, ICME 2005",

}

TY - GEN

T1 - Conditionally positive definite kernels for SVM based image recognition

AU - Boughorbel, Sabri

AU - Tarel, Jean Philippe

AU - Boujemaa, Nozha

PY - 2005

Y1 - 2005

N2 - Kernel based methods such as Support Vector Machine (SVM) have provided successful tools for solving many recognition problems. One of the reason of this success is the use of kernels. Positive defi niteness has to be checked for kernels to be suitable for most of these methods. For instance for SVM, the use of a positive defi nitekernel insures that the optimized problem is convex and thus the obtained solution is unique. Alternative class of kernels called conditionally positive defi nitehave been studied for a long time from the theoretical point of view and have drawn attention from the community only in the last decade. We propose a new kernel, named log kernel, which seems particularly interesting for images. Moreover, we prove that this new kernel is a conditionally positive defi nitekernel as well as the power kernel. Finally, we show from experimentations that using conditionally positive defi nite kernels allows us to outperform classical positive defi nitekernels.

AB - Kernel based methods such as Support Vector Machine (SVM) have provided successful tools for solving many recognition problems. One of the reason of this success is the use of kernels. Positive defi niteness has to be checked for kernels to be suitable for most of these methods. For instance for SVM, the use of a positive defi nitekernel insures that the optimized problem is convex and thus the obtained solution is unique. Alternative class of kernels called conditionally positive defi nitehave been studied for a long time from the theoretical point of view and have drawn attention from the community only in the last decade. We propose a new kernel, named log kernel, which seems particularly interesting for images. Moreover, we prove that this new kernel is a conditionally positive defi nitekernel as well as the power kernel. Finally, we show from experimentations that using conditionally positive defi nite kernels allows us to outperform classical positive defi nitekernels.

UR - http://www.scopus.com/inward/record.url?scp=33750537410&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=33750537410&partnerID=8YFLogxK

U2 - 10.1109/ICME.2005.1521373

DO - 10.1109/ICME.2005.1521373

M3 - Conference contribution

SN - 0780393325

SN - 9780780393325

VL - 2005

SP - 113

EP - 116

BT - IEEE International Conference on Multimedia and Expo, ICME 2005

ER -