Presentation 2016-06-13
Image recognition based on discriminative models using features generated from separable lattice HMMs
Yoshinari Tsuzuki, Kei Sawada, Kei Hashimoto, Yoshihiko Nankaku, Keiichi Tokuda,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) One of the major problems in image recognition is degradation in the recognition performance caused by geometric variation such as position and size of a target object. Separable lattice HMMs (SL-HMMs) have been proposed to solve this problem. However, when using them directly as generative models in classifiers, there is a possibility that the performance is inferior to that of discriminative models, because discriminative models are specialized to identification, while generative models are estimated so as to accurately reproduce training data. In this paper, we propose image recognition based on log linear models (LLMs) using features extracted from SL-HMMs. The proposed technique achieves discriminative model based image recognition invariant to geometric transformation based on derivative features including image normalization in its extraction process. Face recognition experiments showed that the proposed technique obtained higher recognition rates than the generative model based approaches.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Image recognition / hidden Markov hodels / separable lattice HMMs / log linear models / derivative features
Paper # PRMU2016-36,SP2016-2,WIT2016-2
Date of Issue 2016-06-06 (PRMU, SP, WIT)

Conference Information
Committee PRMU / SP / WIT / ASJ-H
Conference Date 2016/6/13(2days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair Eisaku Maeda(NTT) / Kazunori Mano(Shibaura Inst. of Tech.) / Kiyohiko Nunokawa(Tokyo International Univ.)
Vice Chair Seiichi Uchida(Kyushu Univ.) / Hironobu Fujiyoshi(Chubu Univ.) / Hiroki Mori(Utsunomiya Univ.) / Chikamune Wada(Kyushu Inst. of Tech.)
Secretary Seiichi Uchida(Kyoto Univ.) / Hironobu Fujiyoshi(NTT) / Hiroki Mori(Kobe Univ.) / Chikamune Wada(Shizuoka Univ.) / (Nagoya Inst. of Tech.)
Assistant Masaki Oonishi(AIST) / Takuya Funatomi(NAIST) / Taichi Asami(NTT) / Kei Hashimoto(Nagoya Inst. of Tech.) / Tomohiro Amemiya(NTT) / Takeaki Shionome(Tsukuba Univ. of Tech.) / Manabi Miyagi(Tsukuba Univ. of Tech.)

Paper Information
Registration To Technical Committee on Pattern Recognition and Media Understanding / Technical Committee on Speech / Technical Committee on Well-being Information Technology / Auditory Research Meeting
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Image recognition based on discriminative models using features generated from separable lattice HMMs
Sub Title (in English)
Keyword(1) Image recognition
Keyword(2) hidden Markov hodels
Keyword(3) separable lattice HMMs
Keyword(4) log linear models
Keyword(5) derivative features
1st Author's Name Yoshinari Tsuzuki
1st Author's Affiliation Nagoya Institute of Technology(Nagoya Inst. of Tech.)
2nd Author's Name Kei Sawada
2nd Author's Affiliation Nagoya Institute of Technology(Nagoya Inst. of Tech.)
3rd Author's Name Kei Hashimoto
3rd Author's Affiliation Nagoya Institute of Technology(Nagoya Inst. of Tech.)
4th Author's Name Yoshihiko Nankaku
4th Author's Affiliation Nagoya Institute of Technology(Nagoya Inst. of Tech.)
5th Author's Name Keiichi Tokuda
5th Author's Affiliation Nagoya Institute of Technology(Nagoya Inst. of Tech.)
Date 2016-06-13
Paper # PRMU2016-36,SP2016-2,WIT2016-2
Volume (vol) vol.116
Number (no) PRMU-89,SP-90,WIT-91
Page pp.pp.7-12(PRMU), pp.7-12(SP), pp.7-12(WIT),
#Pages 6
Date of Issue 2016-06-06 (PRMU, SP, WIT)