An extension of Separable Lattice 2-D HMMS for rotational data variations

Tamamori,; Nankaku,; Tokuda,

doi:10.1109/icassp.2010.5495735

Cited by 5 publications

(3 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…One of the advantages of SL-HMMs over CNNs is explicit modeling of the generative process, which can represent geometric variations over an entire image. Furthermore, some extensions to structures representing typical geometric variations that are seen in many image recognition tasks have already been proposed, e.g., a structure for rotational variations [18], a structure with multiple horizontal and vertical Markov chains [19], and explicit state duration modeling [20]. By selecting an appropriate model structure reflecting the data generation process for a target task, human knowledge can effectively be utilized as prior information, and this makes it possible to construct models with a small amount of training data.…”

Section: Introductionmentioning

confidence: 99%

A bayesian framework for image recognition based on hidden markov eigen‐image models

Sawada

Hashimoto

Nankaku

et al. 2018

IEEJ Transactions Elec Engng

Self Cite

View full text Add to dashboard Cite

An image recognition method based on hidden Markov eigen‐image models (HMEMs) using a Bayesian framework is proposed and experimentally evaluated. HMEMs have been proposed as a model with two advantageous properties: invariances to the size and location of the object to be recognized, and a linear feature extraction based on statistical analysis. However, the training of HMEMs easily falls into the overfitting problem because HMEMs have a complex model structure, especially when there is insufficient training data. This study aims to accurately estimate HMEMs using the variational Bayesian (VB) method. The VB method can utilize prior distributions representing useful prior information and is expected to have a high generalization ability due to the marginalization of model parameters. Furthermore, to relax the local maximum problem in the VB method, the deterministic annealing expectation maximization algorithm is applied to train HMEMs. Experiments on face recognition indicate that the proposed method offers a significantly improved image recognition performance. Additionally, comparative experimental results show that the proposed method is more robust to geometric variations than convolutional neural networks when the amount of training data is insufficient. © 2018 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

show abstract

Section: Introductionmentioning

confidence: 99%

A bayesian framework for image recognition based on hidden markov eigen‐image models

Sawada

Hashimoto

Nankaku

et al. 2018

IEEJ Transactions Elec Engng

Self Cite

View full text Add to dashboard Cite

show abstract

“…To cope with this problem, the training algorithm for SL2D-HMMs using the variational EM algorithm were derived in [8], where the loglikelihood can be approximated by the variational lower bound. Although some extensions of SL2D-HMMs have been proposed, e.g., a structure for rotational variations [10], explicit state duration modeling [11], and a structure with multiple horizontal/vertical Markov chains [12], this paper uses an original form of SL2D-HMMs.…”

Section: Separable Lattice 2-d Hmmsmentioning

confidence: 99%

Image Recognition Based on Separable Lattice Trajectory 2-D HMMs

Tamamori

Nankaku

Tokuda

2014

IEICE Trans. Inf. & Syst.

Self Cite

View full text Add to dashboard Cite

Akira TAMAMORI†a) , Nonmember, Yoshihiko NANKAKU †b) , and Keiichi TOKUDA †c) , Members SUMMARYIn this paper, a novel statistical model based on 2-D HMMs for image recognition is proposed. Recently, separable lattice 2-D HMMs (SL2D-HMMs) were proposed to model invariance to size and location deformation. However, their modeling accuracy is still insufficient because of the following two assumptions, which are inherited from 1-D HMMs: i) the stationary statistics within each state and ii) the conditional independent assumption of state output probabilities. To overcome these shortcomings in 1-D HMMs, trajectory HMMs were proposed and successfully applied to speech recognition and speech synthesis. This paper derives 2-D trajectory HMMs by reformulating the likelihood of SL2D-HMMs through the imposition of explicit relationships between static and dynamic features. The proposed model can efficiently capture dependencies between adjacent observations without increasing the number of model parameters. The effectiveness of the proposed model was evaluated in face recognition experiments on the XM2VTS database.

show abstract

“…One of advantages of SLHMMs against CNNs is explicit modeling of generative process which can represent geometric variations over an entire image. Furthermore, some extensions to structures representing typical geometric variations which are seen in many Copyright c 2016 The Institute of Electronics, Information and Communication Engineers image recognition tasks have already been proposed, e.g., a structure for rotational variations [16], a structure with multiple horizontal and vertical Markov chains [17], explicit state duration modeling [18], trajectory modeling [19], and integration SL-HMMs and factor analyzers [20]. By selecting an appropriate model structures reflecting data generation process for a target task, human knowledge can effectively be utilized as prior information and this makes it possible to construct classifiers with a small amount of training data.…”

Section: Introductionmentioning

confidence: 99%

A Bayesian Approach to Image Recognition Based on Separable Lattice Hidden Markov Models

Sawada

Tamamori

Hashimoto

et al. 2016

IEICE Trans. Inf. & Syst.

Self Cite

View full text Add to dashboard Cite

SUMMARYThis paper proposes a Bayesian approach to image recognition based on separable lattice hidden Markov models (SL-HMMs). The geometric variations of the object to be recognized, e.g., size, location, and rotation, are an essential problem in image recognition. SL-HMMs, which have been proposed to reduce the effect of geometric variations, can perform elastic matching both horizontally and vertically. This makes it possible to model not only invariances to the size and location of the object but also nonlinear warping in both dimensions. The maximum likelihood (ML) method has been used in training SL-HMMs. However, in some image recognition tasks, it is difficult to acquire sufficient training data, and the ML method suffers from the over-fitting problem when there is insufficient training data. This study aims to accurately estimate SL-HMMs using the maximum a posteriori (MAP) and variational Bayesian (VB) methods. The MAP and VB methods can utilize prior distributions representing useful prior information, and the VB method is expected to obtain high generalization ability by marginalization of model parameters. Furthermore, to overcome the local maximum problem in the MAP and VB methods, the deterministic annealing expectation maximization algorithm is applied for training SL-HMMs. Face recognition experiments performed on the XM2VTS database indicated that the proposed method offers significantly improved image recognition performance. Additionally, comparative experiment results showed that the proposed method was more robust to geometric variations than convolutional neural networks. key words: image recognition, hidden Markov models, separable lattice hidden Markov models, Bayesian approach, deterministic annealing

show abstract

An extension of Separable Lattice 2-D HMMS for rotational data variations

Cited by 5 publications

References 4 publications

A bayesian framework for image recognition based on hidden markov eigen‐image models

A bayesian framework for image recognition based on hidden markov eigen‐image models

Image Recognition Based on Separable Lattice Trajectory 2-D HMMs

A Bayesian Approach to Image Recognition Based on Separable Lattice Hidden Markov Models

Contact Info

Product

Resources

About