Tetsunori Kobayashi

Subspace-based age-group classification using facial images under various lighting conditions

This paper presents a framework of age-group classification using facial images under various lighting conditions. Our method is based on the appearance-based approach that projects images from the original image space into a face-subspace. We propose a two-phased approach (2DLDA+LDA), which is based on 2DPCA and LDA. Our experimental results show that the new 2DLDA+LDA-based approach improves classification accuracy more than the conventional PCA-based and LDA-based approach. Moreover, the effectiveness of eliminating dimensions that do not contain important discriminative information is confirmed. The accuracy rates are 46.3%, 67.8% and 78.1% for age-groups that are in the 5-year, 10-year and 15-year range respectively

A conversation robot using head gesture recognition as para-linguistic information

A conversation robot that recognizes users head gestures and uses its results as para-linguistic information is developed. In the conversation, humans exchange linguistic information, which can be obtained by transcription of the utterance, and para-linguistic information, which helps the transmission of linguistic information. Para-linguistic information brings a nuance that cannot be transmitted by linguistic information, and the natural and effective conversation is realized. We recognize users head gestures as the para-linguistic information in the visual channel. We use the optical flow over the head region as the feature and model them using HMM for the recognition. In actual conversation, while the user performs a gesture, the robot may perform a gesture, too. In this situation, the image sequence captured by the camera mounted on the eyes of the robot includes sways caused by the movement of the camera. To solve this problem, we introduced two artifices. One is for the feature extraction: the optical flow of the body area is used to compensate the swayed images. The other is for the probability models: mode-dependent models are prepared by the MLLR model adaptation technique, and the models are switched according to the motion mode of the robot. Experimental results show the effectiveness of these techniques.

Partly-hidden Markov model and its application to gesture recognition

A new pattern matching method, the partly-hidden Markov model, is proposed for gesture recognition. The hidden Markov model, which is widely used for the time series pattern recognition, can deal with only piecewise stationary stochastic process. We solved this problem by introducing the modified second order Markov model, in which the first state is hidden and the second one is observable. As shown by the results of 6 sign-language recognition test, the error rate was improved by 73% compared with normal HMM.

Estimating articulatory motion from speech wave

Abstract If articulatory movements can be estimated, then the articulatory parameters which represent the motion of the articulatory organs would be useful for speech recognition. This paper discusses an effective method of estimating articulatory movements and its application to speech recognition. Firstly, what is described is a method of estimating articulatory parameters known as the model matching method, and various spectral distance measures are evaluated for this method. The results show that the best in average is the higher order cepstral distance measure, which is one of the peak weighted measure. Secondly, articulatory parameters are utilized for the recognition of vowels uttered by unspecified speakers. It is shown that the adaptation of the model by the estimated mean vocal tract length is effective to normalize speaker difference. Thirdly, the motor commands to move the articulatory organs are estimated considering articulatory dynamics, and the continuous vowels are recognized by means of these estimated commands. It has been found that a considerable part of the coarticulation effects can be compensated for by this command estimated, and the method is useful for continuous speech recognition.

A method of gender classification by integrating facial, hairstyle, and clothing images

This work presents a method of gender classification by integrating facial, hairstyle, and clothing images. Initially, input images are separated into facial, hairstyle and clothing regions, and independently learned PCAs and GMMs based on thousands of sample images are applied to each region. The classification results are then integrated into a single score using some known priors based on the Bayes rule. Experimental results showed that our integration strategy significantly reduced error rate in gender classification compared with the conventional facial only approach.

Extension of hidden Markov models to deal with multiple candidates of observations and its application to mobile-robot-oriented gesture recognition

We propose a modified hidden Markov model (HMM) with a view to improving gesture recognition in the moving camera condition. We define a new gesture recognition framework in which multiple candidates of feature vectors are generated with confidence measures and the HMM is extended to deal with these multiple feature vectors. Experimental analysis comparing the proposed system with feature vectors based on DCT and the method of selecting only one candidate feature point verifies the effectiveness of the technique.

Application of neural networks to articulatory motion estimation

The authors discuss an application of neural networks (NNs) to the problem of estimating the motion of articulatory organs from speech waves. A four-layer feedforward network was successfully applied to the articulatory parameter estimation problem. The evaluation test was performed using the vowel data in 5200 tokens in the ATR word database. Results show that the difference in estimated articulatory parameter values between the conventional model matching method (MM) and NN is only 0.1, which is about 3% of the value range, on average. For a few data, large differences arise between MM and NN, but this is due to misestimation in MM rather than NN. The percentage of misestimates in NN is less than 50% of that for MM. As for calculation time, NN is 10 times faster than MM. Thus, a high-speed and stable articulatory parameter estimation technique can be realized using neural networks.<<ETX>>

Controlling gaze of humanoid in communication with human

This paper describes controlling robots gaze which has relation to smoothness of turn-taking in communication. We considered the role of gaze in dialogues between human beings and examined it by simulation and our humanoid. Also we analyzed the features of gaze movement in dialogues by plural persons and confirmed that controlling gaze is efficient in confirmation of communication channel by implementing it on the humanoid.

Four-participant group conversation: A facilitation robot controlling engagement density as the fourth participant

Abstract In this paper, we present a framework for facilitation robots that regulate imbalanced engagement density in a four-participant conversation as the forth participant with proper procedures for obtaining initiatives. Four is the special number in multiparty conversations. In three-participant conversations, the minimum unit for multiparty conversations, social imbalance, in which a participant is left behind in the current conversation, sometimes occurs. In such scenarios, a conversational robot has the potential to objectively observe and control situations as the fourth participant. Consequently, we present model procedures for obtaining conversational initiatives in incremental steps to harmonize such four-participant conversations. During the procedures, a facilitator must be aware of both the presence of dominant participants leading the current conversation and the status of any participant that is left behind. We model and optimize these situations and procedures as a partially observable Markov decision process (POMDP), which is suitable for real-world sequential decision processes. The results of experiments conducted to evaluate the proposed procedures show evidence of their acceptability and feeling of groupness.

Media-integrated biometric person recognition based on the Dempster-Shafer theory

The paper describes a new integration method of speech and facial image information for person recognition problems based on the Dempster-Shafer probability theory. The Dempster-Shafer theory provides an attractive methodology by which to integrate multiple numerical evidences containing ambiguities. However, no concrete and reasonable methodology exists to enumerate the reliability of evidences. In the present paper, this problem is solved using the cumulative density function of both the correct and incorrect categories. The proposed enumerating method allows the Dempster-Shafer theory to be applied to media integration. We show that the total performance of person recognition, including rejection of unregistered users, is improved significantly using the proposed method.