• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.37 no.6
• /
• pp.22-31
• /
• 2000

In this paper, we propose a method for removing noise components in the feature extraction process for robust speech recognition. This method is based on blind separation using independent component analysis (ICA). Given two noisy speech recordings the algorithm linearly separates speech from the unwanted noise signal. To apply ICA as closely as possible to the feature level for recognition, a new spectral analysis is presented. It modifies the computation of band energies by previously averaging out fast Fourier transform (FFT) points in several divided ranges within one met-scaled band. The simple analysis using sample variances of band energies of speech and noise, and recognition experiments showed its noise robustness. For noisy speech signals recorded in real environments, the proposed method which applies ICA to the new spectral analysis improved the recognition performances to a considerable extent, and was particularly effective for low signal-to-noise ratios (SNRs). This method gives some insights into applying ICA to feature levels and appears useful for robust speech recognition.

• Journal of KIISE:Software and Applications
• /
• v.29 no.8
• /
• pp.573-583
• /
• 2002

Pattern recognition process is made up of the feature extraction in the pre-processing, the pattern clustering by training and the recognition process. This paper presents the F-LVQ (Fuzzy Learning Vector Quantization) pattern classification model which is optimized by the fuzzy membership function for the OCR(Optical Character Recognition) system. We trained 220 numeric patterns of 22 Hangul and English fonts and tested 4840 patterns whose forms are changed variously. As a result of this experiment, it is proved that the proposed model is more effective and robust than other typical LVQ models.

• 전자공학회논문지 IE
• /
• v.43 no.4
• /
• pp.37-42
• /
• 2006

This paper proposes a adaptive video watermarking algorithm according to bitrate and motion vector size in MPEG2 system. The watermark strength in the I-frames is adapted for quantization step size and the strength in the P-B-frames is adapted for quantization step size and motion vector of macroblock to make the watermark more robust against the accompanying degradation due to aggressively compression. A realtime watermark extraction is done directly in the DCT domain during MPEG decoding without full decoding of MPEG video. The experimental simulations show that the video quality results almost invisible difference between the watermarked frames and the original frames and the watermark is resistant to frame dropping, MPEG compression, GoP conversion and low pass filter attacks.

• Journal of Biomedical Engineering Research
• /
• v.33 no.2
• /
• pp.65-71
• /
• 2012

In this paper, the Gaussian Mixture Model(GMM) which is very robust modeling for pattern classification is proposed to classify wrist motions using surface electromyograms(EMG). EMG is widely used to recognize wrist motions such as up, down, left, right, rest, and is obtained from two electrodes placed on the flexor carpi ulnaris and extensor carpi ulnaris of 15 subjects under no strain condition during wrist motions. Also, EMG-based feature is derived from extracted EMG signals in time domain for fast processing. The estimated features based in difference absolute mean value(DAMV) are used for motion classification through GMM. The performance of our approach is evaluated by recognition rates and it is found that the proposed GMM-based method yields better results than conventional schemes including k-Nearest Neighbor(k-NN), Quadratic Discriminant Analysis(QDA) and Linear Discriminant Analysis(LDA).

• Journal of Institute of Control, Robotics and Systems
• /
• v.19 no.1
• /
• pp.15-26
• /
• 2013

This paper proposes a classification method of parallel, vertical parking states and pillars for parking assist system using ultrasonic sensors. Since, in general parking space detection module, the compressed amplitude of ultrasonic data are received, the analysis of them is difficult. To solve these problems, in preprocessing state, symmetric transform and noise removal are performed. In feature extraction process, four features, standard deviation of distance, reconstructed peak, standard deviation of reconstructed signal and sum of width, are proposed. Gaussian fitting model is used to reconstruct saturated peak signal and discriminability of each feature is measured. To find the best combination among these features, multi-class SVM and subset generator are used for more accurate and robust classification. The proposed method shows 92 % classification rate and proves the applicability to parking space detection modules.

• Journal of Institute of Control, Robotics and Systems
• /
• v.17 no.1
• /
• pp.22-30
• /
• 2011

Plane detection is very important information for mission-critical of robot in 3D environment. A representative method of plane detection is Hough-transformation. Hough-transformation is robust to noise and makes the accurate plane detection possible. But it demands excessive memory and takes too much processing time. Iterative randomized Hough-transformation has been proposed to overcome these shortcomings. This method doesn't vote all data. It votes only one value of the randomly selected data into the Hough parameter space. This value calculated the value of the parameter of the shape that we want to extract. In Hough parameters space, it is possible to detect accurate plane through detection of repetitive maximum value. A common problem in these methods is that it requires too much computational cost and large number of memory space to find the distribution of mixed multiple planes in parameter space. In this paper, we detect multiple planes only via data sampling using Self Organizing Map method. It does not use conventional methods that include transforming to Hough parameter space, voting and repetitive plane extraction. And it improves the reliability of plane detection through division area searching and planarity evaluation. The proposed method is more accurate and faster than the conventional methods which is demonstrated the experiments in various conditions.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.2
• /
• pp.907-923
• /
• 2020

For the severe haze situation in the Beijing-Tianjin-Hebei region, conventional fine particulate matter (PM2.5) concentration prediction methods based on pollutant data face problems such as incomplete data, which may lead to poor prediction performance. Therefore, this paper proposes a method of predicting the PM2.5 concentration based on image analysis technology that combines image data, which can reflect the original weather conditions, with currently popular machine learning methods. First, based on local parameter estimation, autoregressive (AR) model analysis and local estimation of the increase in image blur, we extract features from the weather images using an approach inspired by free energy and a no-reference robust metric model. Next, we compare the coefficient energy and contrast difference of each pixel in the AR model and then use the percentages to calculate the image sharpness to derive the overall mass fraction. Furthermore, the results are compared. The relationship between residual value and PM2.5 concentration is fitted by generalized Gauss distribution (GGD) model. Finally, nonlinear mapping is performed via the wavelet neural network (WNN) method to obtain the PM2.5 concentration. Experimental results obtained on real data show that the proposed method offers an improved prediction accuracy and lower root mean square error (RMSE).

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.6
• /
• pp.2709-2729
• /
• 2016

Representation based classification, kernel method and sparse representation have received much attention in the field of face recognition. In this paper, we proposed an improved kernel principal component analysis method based on sparse representation to improve the accuracy and robustness for face recognition. First, the distances between the test sample and all training samples in kernel space are estimated based on collaborative representation. Second, S training samples with the smallest distances are selected, and Kernel Principal Component Analysis (KPCA) is used to extract the features that are exploited for classification. The proposed method implements the sparse representation under ℓ₂ regularization and performs feature extraction twice to improve the robustness. Also, we investigate the relationship between the accuracy and the sparseness coefficient, the relationship between the accuracy and the dimensionality respectively. The comparative experiments are conducted on the ORL, the GT and the UMIST face database. The experimental results show that the proposed method is more effective and robust than several state-of-the-art methods including Sparse Representation based Classification (SRC), Collaborative Representation based Classification (CRC), KCRC and Two Phase Test samples Sparse Representation (TPTSR).

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.9 no.4
• /
• pp.343-350
• /
• 2016

In this paper, We proposed a realtime smoke detection using hidden markov model and DWT. The smoke type is not clear. The color of the smoke, form, spread direction, etc., are characterized by varying the environment. Therefore, smoke detection using specific information has a high error rate detection. Dynamic Object Detection was used a robust foreground extraction method to environmental changes. Smoke recognition is used to integrate the color, shape, DWT energy information of the detected object. The proposed method is a real-time processing by having the average processing speed of 30fps. The average detection time is about 7 seconds, it is possible to detect early rapid.

• The KIPS Transactions:PartB
• /
• v.16B no.4
• /
• pp.257-262
• /
• 2009

A method of the region extraction and recognition of a speed sign in the real road environment is proposed. The region of speed sign is extracted by using color information and then numbers are segmented in the region. We improve the recognition rate by performing an incline compensation of the speed sign for directions clockwise and counterclockwise. In image sequences of the real road environment, a robust recognition results are achieved with speed signs at normal condition as well as inclined.