• The Journal of the Acoustical Society of Korea
• /
• 제18권3E호
• /
• pp.3-8
• /
• 1999

In speech recognition for real-world applications, the performance degradation due to the mismatch introduced between training and testing environments should be overcome. In this paper, to reduce this mismatch, we provide a hybrid method of spectral subtraction and residual noise masking. We also employ multiple model approach to obtain improved robustness over various noise environments. In this approach, multiple model sets are made according to several noise masking levels and then a model set appropriate for the estimated noise level is selected automatically in recognition phase. According to speaker independent isolated word recognition experiments in car noise environments, the proposed method using model sets with only two masking levels reduced average word error rate by 60% in comparison with spectral subtraction method.

• 인지과학
• /
• 제17권2호
• /
• pp.139-150
• /
• 2006

음성 대화 시스템에서 음성 인식 오류는 전체 시스템의 치명적인 결과를 초래한다. 음성 인식 오류가 부분적으로 발생하여 화행 분석이 실패했을 때 시스템은 원활한 대화를 진행할 수 없다. 본 논문은 양식 채우기 대화 형식에서 발생하는 음성 인식 오류 유형에 따라 시스템이 사용자 발화의 화행을 추론하기 위한 부대화 생성 전략을 제안한다. 제안하는 방법을 계획기반 대화 모델로 구현하여 실험하였고, 사용자 작업 실패 오류의 약27%를 보완하여 성능을 향상시켰으며 전체 시스템의 사용자 작업 성공률은 약 89%이다.

• 산업기술연구
• /
• 제28권A호
• /
• pp.185-189
• /
• 2008

Practical pattern recognition systems should overcome very large class problem. Sometimes it is almost impossible to build every model for every class due to memory and time constraints. For this case, grouping similar models will be helpful. In this paper, we propose GMC(Group Model Clustering) to build a large class Chinese character recognition system. We built hidden Markov models for 10% of total classes, then classify the rest of classes into already trained group classes. Finally group models are trained using group model clustered data. Recognition is performed using only group models, in order to achieve reduced model size and improved recognition speed.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제5권4호
• /
• pp.782-804
• /
• 2011

In video-based recognition systems, stationary cameras are used to monitor an area of interest. These systems focus on a segmentation of the foreground in the video stream and the recognition of the events occurring in that area. The usual approach to discriminating the foreground from the video sequence is background subtraction. This paper presents a novel background subtraction method based on a probabilistic approach. We represent the posterior probability of the foreground based on the current image and all past images and derive an updated method. Furthermore, we present an efficient fusion method for the color and edge information in order to overcome the difficulties of existing background subtraction methods that use only color information. The suggested method is applied to synthetic data and real video streams, and its robust performance is demonstrated through experimentation.

• Journal of Electrical Engineering and Technology
• /
• 제10권2호
• /
• pp.659-669
• /
• 2015

Face Recognition assumes much significance in the context of security based application. Normally, high resolution images offer more details about the image and recognizing a face from a reasonably high resolution image would be easier when compared to recognizing images from very low resolution images. This paper addresses the problem of recognizing faces from a very low resolution image whose size is as low as $8{\times}8$. With the use of CCTV(Closed Circuit Television) and with other surveillance camera-based application for security purposes, the need to overcome the shortcomings with very low resolution images has been on the rise. The present day face recognition algorithms could not provide adequate performance when employed to recognize images from VLR images. Existing methods use super-resolution (SR) methods and Relation Based Super Resolution methods to construct from very low resolution images. This paper uses a learning based super resolution method to extract and construct images from very low resolution images. Experimental results show that the proposed SR algorithm based on relationship learning outperforms the existing algorithms in public face databases.

• Journal of Multimedia Information System
• /
• 제5권3호
• /
• pp.179-188
• /
• 2018

In order to overcome the problems of contact palmprint recognition, a non-contact palmprint recognition system is developed on personal computer (PC) platform. Three methods, namely "double-line-single-point" (DLSP), "double-assistant-crosshair" (DAC) and "none-assistant-graphic" (NAG), are implemented for the palmprint localization to solve the severe technical challenges, including the complex background, variant illuminations, uncontrollable locations and gestures of hands. In NAG, hand segmentation and the cropping of region of interest are performed without any assistant graphics. The convex hull contour of hand helps detect the outside contour of little finger as well as the valley bottom between thumb and index finger. The three methods of palmprint localization have good operating efficiency and can meet the performance requirements of real-time system. Furthermore, an attendance system on PC platform is designed and developed based on non-contact palmprint recognition.

• The Journal of the Acoustical Society of Korea
• /
• 제23권2E호
• /
• pp.51-55
• /
• 2004

In this paper, we propose a style-specific language model adaptation scheme using n-gram based tf*idf similarity for Korean spontaneous speech recognition. Korean spontaneous speech shows especially different style-specific characteristics such as filled pauses, word omission, and contraction, which are related to function words and depend on preceding or following words. To reflect these style-specific characteristics and overcome insufficient data for training language model, we estimate in-domain dependent n-gram model by relevance weighting of out-of-domain text data according to their n-. gram based tf*idf similarity, in which in-domain language model include disfluency model. Recognition results show that n-gram based tf*idf similarity weighting effectively reflects style difference.

• 한국통신학회논문지
• /
• 제15권6호
• /
• pp.455-466
• /
• 1990

영문자 및 기호를 인식할 수 있는 프로그램을 개발하였으며, 이 소프트웨어로 전처리 수행한 결과 속도의 한계성이 있었다. 이 속도의 한계를 극복하고자 신경회로망 알고리즘을 이용해 전처리 과정용 집적회로 칩을 설계하였다. 설계된 칩은 잡음제거, 선형화, 세선화 및 특징점 추출을 위한 것이다. 이 칩들은 단층 구조 퍼셉트론 신경회로 모델에 따라 CMOS 이중 금속 2um 설계 규칙에 의하 설계되었다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제10권2호
• /
• pp.787-806
• /
• 2016

Traditional radio-based gesture recognition approaches usually require the target to carry a device (e.g., an EMG sensor or an accelerometer sensor). However, such requirement cannot be satisfied in many applications. For example, in smart home, users want to control the light on/off by some specific hand gesture, without finding and pressing the button especially in dark area. They will not carry any device in this scenario. To overcome this drawback, in this paper, we propose three algorithms able to recognize the target gesture (mainly the human hand gesture) without carrying any device, based on just Radio Signal Strength Indicator (RSSI). Our platform utilizes only 6 telosB sensor nodes with a very easy deployment. Experiment results show that the successful recognition radio can reach around 80% in our system.

• 반도체디스플레이기술학회지
• /
• 제19권2호
• /
• pp.96-99
• /
• 2020

According to brilliant development of smart devices, many related services are being devised. And, almost every service is designed to provide user-centric services based on personal information. In this situation, to prevent unintentional leakage of personal information is essential. Conventionally, ID and Password system is used for the user authentication. This is a convenient method, but it has a vulnerability that can cause problems due to information leakage. To overcome these problem, many methods related to face recognition is being researched. Through this paper, we investigated the trend of user authentication through biometrics and a representative model for face recognition techniques. One is DeepFace of FaceBook and another is FaceNet of Google. Each model is based on the concept of Deep Learning and Distance Metric Learning, respectively. And also, they are based on Convolutional Neural Network (CNN) model. In the future, further research is needed on the equipment configuration requirements for practical applications and ways to provide actual personalized services.