• ETRI Journal
• /
• v.24 no.6
• /
• pp.469-472
• /
• 2002

Speaker adaptation techniques are generally used to reduce speaker differences in speech recognition. In this work, we focus on the features fitted to a linear regression-based speaker adaptation. These are obtained by feature transformation based on independent component analysis (ICA), and the feature transformation matrices are estimated from the training data and adaptation data. Since the adaptation data is not sufficient to reliably estimate the ICA-based feature transformation matrix, it is necessary to adjust the ICA-based feature transformation matrix estimated from a new speaker utterance. To cope with this problem, we propose a smoothing method through a linear interpolation between the speaker-independent (SI) feature transformation matrix and the speaker-dependent (SD) feature transformation matrix. From our experiments, we observed that the proposed method is more effective in the mismatched case. In the mismatched case, the adaptation performance is improved because the smoothed feature transformation matrix makes speaker adaptation using noisy speech more robust.

• Journal of IKEEE
• /
• v.24 no.2
• /
• pp.419-425
• /
• 2020

The market size of immersive 360-degree video contents, which are noted as one of the main technology of the fourth industry, increases every year. However, since most of the images are distributed through illegal distribution networks such as Torrent after the DRM gets lifted, the damage caused by illegal copying is also increasing. Although filtering technology is used as a technology to respond to these issues in 2D videos, most of those filtering technology has issues in that it has to overcome the technical limitation such as huge feature-point data volume and the related processing capacity due to ultra high resolution such as 4K UHD or higher in order to apply the existing technology to immersive 360° videos. To solve these problems, this paper proposes a feature-point recognition ratio improvement method for immersive 360-degree videos using deep learning technology.

• Proceedings of the Korean Society of Machine Tool Engineers Conference
• /
• 2004.04a
• /
• pp.287-292
• /
• 2004

In this study, we made a comparative study of backpropagation neural network and probabilistic neural network and bayesian classifier and perceptron as shape recognition algorithm of welding flaws. For this purpose, variables are applied the same to four algorithms. Here, feature variable is composed of time domain signal itself and frequency domain signal itself, Through this process, we confirmed advantages/disadvantages of four algorithms and identified application methods of few algorithms.

• Proceedings of the Korean Society of Machine Tool Engineers Conference
• /
• 2003.10a
• /
• pp.271-276
• /
• 2003

In this study, we compared backpropagation neural network(BPNN) with probabilistic neural network(PNN) as shape recognition algorithm of welding flaws. For this purpose, variables are applied the same to two algorithm. Here, feature variable is composed of time domain signal itself and frequency domain signal itself, Through this process, we comfirmed advantages/disadvantages of two algorithms and identified application methods of two algorithms.

• Proceedings of the KSPS conference
• /
• 2005.04a
• /
• pp.103-106
• /
• 2005

The goal of our research is to build a text independent speaker recognition system that can be used in any condition without any additional adaptation process. The performance of speaker recognition systems can be severally degraded in some unknown mismatched microphone and noise conditions. In this paper, we show that PCA(Principal component analysis) without dimension reduction can greatly increase the performance to a level close to matched condition. The error rate is reduced more by the proposed augmented PCA, which augment an axis to the feature vectors of the most confusable pairs of speakers before PCA

• Journal of Information Management
• /
• v.41 no.2
• /
• pp.31-46
• /
• 2010

Now named-entity recognition(NER) as a part of information extraction has been used in the fields of information retrieval as well as question-answering systems. Unlike words, named-entities(NEs) are generated and changed steadily in documents on the Web, newspapers, and so on. The NE generation causes an unknown word problem and makes many application systems with NER difficult. In order to alleviate this problem, this paper proposes a new feature generation method for machine learning-based NER. In general features in machine learning-based NER are related with words, but entities in named-entity dictionaries are related to phrases. So the entities are not able to be directly used as features of the NER systems. This paper proposes an encoding scheme as a feature generation method which converts phrase entities into features of word units. Futhermore, due to this scheme, entities with semantic information in WordNet can be converted into features of the NER systems. Through our experiments we have shown that the performance is increased by about 6% of F1 score and the errors is reduced by about 38%.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.11
• /
• pp.4153-4169
• /
• 2014

To recognize F5-like (such as F5 and nsF5) steganographic algorithm from multi-class stego images, a recognition algorithm based on the identifiable statistical feature (IDSF) of F5-like steganography is proposed in this paper. First, this paper analyzes the special modification ways of F5-like steganography to image data, as well as the special changes of statistical properties of image data caused by the modifications. And then, by constructing appropriate feature extraction sources, the IDSF of F5-like steganography distinguished from others is extracted. Lastly, based on the extracted IDSFs and combined with the training of SVM (Support Vector Machine) classifier, a recognition algorithm is presented to recognize F5-like stego images from images set consisting of a large number of multi-class stego images. A series of experimental results based on the detection of five types of typical JPEG steganography (namely F5, nsF5, JSteg, Steghide and Outguess) indicate that, the proposed algorithm can distinguish F5-like stego images reliably from multi-class stego images generated by the steganography mentioned above. Furthermore, even if the types of some detected stego images are unknown, the proposed algorithm can still recognize F5-like stego images correctly with high accuracy.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2018.10a
• /
• pp.556-557
• /
• 2018

About 350 shelters nationwide go through about 100,000 organic animals every year. If you are not adopted by the adoption candidate, you will be euthanized in just one out of every fourteen days after entering the shelter. Therefore, in order to prevent the occurrence of organic animals, it is necessary to register the companion animal easily and to register the inscription to manage the history. In this paper, we propose a local technician who can describe feature points in inscription images to develop recognition technology through inscription, which can distinguish companion animals such as human fingerprints.

• Journal of Electrical Engineering and Technology
• /
• v.12 no.1
• /
• pp.363-372
• /
• 2017

This paper presents a novel face recognition algorithm based on the deep convolution neural network and key point detection jointed local binary pattern methodology to enhance the accuracy of face recognition. We firstly propose the modified face key feature point location detection method to enhance the traditional localization algorithm to better pre-process the original face images. We put forward the grey information and the color information with combination of a composite model of local information. Then, we optimize the multi-layer network structure deep learning algorithm using the Fisher criterion as reference to adjust the network structure more accurately. Furthermore, we modify the local binary pattern texture description operator and combine it with the neural network to overcome drawbacks that deep neural network could not learn to face image and the local characteristics. Simulation results demonstrate that the proposed algorithm obtains stronger robustness and feasibility compared with the other state-of-the-art algorithms. The proposed algorithm also provides the novel paradigm for the application of deep learning in the field of face recognition which sets the milestone for further research.

• Journal of Korea Society of Industrial Information Systems
• /
• v.17 no.3
• /
• pp.1-7
• /
• 2012

As personal information is utilized as an important user authentication means, a trustable certification means is being required. Recently, a research on the biometrics system using a part of the human body like a password is being attempted a lot. The face recognition technology using characteristics of the personal face among several biometrics technologies is easy in extracting features. In this paper, we implement a face recognition system for unattended reception interface. Our method is performed by two steps. Firstly the face is extracted using Haar-like feature method. Secondly the method combining PCA and LDA for face recognition was used. To assess the effectiveness of the proposed system, it was tested and experimental results show that the proposed method is applicable for unattended reception interface.