• Journal of the Institute of Convergence Signal Processing
• /
• v.14 no.4
• /
• pp.235-242
• /
• 2013

In this paper, the system which can translate sign language through motion recognition of Kinect camera system is developed for the communication between hearing-impaired person or language disability, and normal person. The proposed algorithm which can translate sign language is developed by using core function of Kinect, and two ways such as length normalization and elbow normalization are introduced to improve accuracy of translating sign langauge for various sign language users. After that the sign language data is compared by chart in order to know how effective these ways of normalization. The accuracy of this program is demonstrated by entering 10 databases and translating sign languages ranging from simple signs to complex signs. In addition, the reliability of translating sign language is improved by applying this program to people who have various body shapes and fixing measure errors in body shapes.

• Journal of the Institute of Convergence Signal Processing
• /
• v.3 no.1
• /
• pp.8-14
• /
• 2002

In this paper, we realized a real-time text-independent speaker recognition system using gaussian mixture model, and applied frame level likelihood normalization method which shows its effects in verification system. The system has three parts as front-end, training, recognition. In front-end part, cepstral mean normalization and silence removal method were applied to consider speaker's speaking variations. In training, gaussian mixture model was used for speaker's acoustic feature modeling, and maximum likelihood estimation was used for GMM parameter optimization. In recognition, likelihood score was calculated with speaker models and test data at frame level. As test sentences, we used text-independent sentences. ETRI 445 and KLE 452 database were used for training and test, and cepstrum coefficient and regressive coefficient were used as feature parameters. The experiment results show that the frame-level likelihood method's recognition result is higher than conventional method's, independently the number of registered speakers.

• Korean Journal of Remote Sensing
• /
• v.28 no.4
• /
• pp.393-407
• /
• 2012

This study focuses on the correction of topographic effects caused by a combination of solar elevation and azimuth, and topographic relief in single optical remote sensing imagery, and by a combination of changes in position of the sun and topographic relief in comparative analysis of multi-temporal imageries. For the Jeju Island, Republic of Korea, where Mt. Halla and various cinder cones are located, a Landsat 7 ETM+ imagery and ASTER GDEM data were used to normalize the topographic effects on the imagery, using two topographic normalization methods: cosine correction assuming a Lambertian condition and assuming a non-Lambertian c-correction, with kernel sizes of $3{\times}3$, $5{\times}5$, $7{\times}7$, and $9{\times}9$ pixels. The effects of each correction method and kernel size were then evaluated. The c-correction with a kernel size of $7{\times}7$ produced the best result in the case of a land area with various land-cover types. For a land-cover type of forest extracted from an unsupervised classification result using the ISODATA method, the c-correction with a kernel size of $9{\times}9$ produced the best result, and this topographic normalization for a single land cover type yielded better compensation for topographic effects than in the case of an area with various land-cover types. In applying the relative radiometric normalization to topographically normalized three multi-temporal imageries, more invariant spectral reflectance was obtained for infrared bands and the spectral reflectance patterns were preserved in visible bands, compared with un-normalized imageries. The results show that c-correction considering the remaining reflectance energy from adjacent topography or imperfect atmospheric correction yielded superior normalization results than cosine correction. The normalization results were also improved by increasing the kernel size to compensate for vertical and horizontal errors, and for displacement between satellite imagery and ASTER GDEM.

• Journal of KIISE:Databases
• /
• v.33 no.6
• /
• pp.600-619
• /
• 2006

According to recent technical advances on sensors and mobile devices, processing of data streams generated by the devices is becoming an important research issue. The data stream of real values obtained at continuous time points is called streaming time-series. Due to the unique features of streaming time-series that are different from those of traditional time-series, similarity matching problem on the streaming time-series should be solved in a new way. In this paper, we propose an efficient algorithm for streaming time- series matching problem that supports normalization transform. While the existing algorithms compare streaming time-series without any transform, the algorithm proposed in the paper compares them after they are normalization-transformed. The normalization transform is useful for finding time-series that have similar fluctuation trends even though they consist of distant element values. The major contributions of this paper are as follows. (1) By using a theorem presented in the context of subsequence matching that supports normalization transform[4], we propose a simple algorithm for solving the problem. (2) For improving search performance, we extend the simple algorithm to use $k\;({\geq}\;1)$ indexes. (3) For a given k, for achieving optimal search performance of the extended algorithm, we present an approximation method for choosing k window sizes to construct k indexes. (4) Based on the notion of continuity[8] on streaming time-series, we further extend our algorithm so that it can simultaneously obtain the search results for $m\;({\geq}\;1)$ time points from present $t_0$ to a time point $(t_0+m-1)$ in the near future by retrieving the index only once. (5) Through a series of experiments, we compare search performances of the algorithms proposed in this paper, and show their performance trends according to k and m values. To the best of our knowledge, since there has been no algorithm that solves the same problem presented in this paper, we compare search performances of our algorithms with the sequential scan algorithm. The experiment result showed that our algorithms outperformed the sequential scan algorithm by up to 13.2 times. The performances of our algorithms should be more improved, as k is increased.

• Structural Monitoring and Maintenance
• /
• v.9 no.2
• /
• pp.179-200
• /
• 2022

This paper proposes a data-driven methodology for online early damage identification under changing environmental conditions. The proposed method relies on two data analysis methods: feature-based method and hybrid principal component analysis (PCA) and kernel PCA to separate damage from environmental influences. First, spectral sub-band features, namely, spectral sub-band centroids (SSCs) and log spectral sub-band energies (LSSEs), are proposed as damage-sensitive features to extract damage information from measured structural responses. Second, hybrid modeling by integrating PCA and kernel PCA is performed on the spectral sub-band feature matrix for data normalization to extract both linear and nonlinear features for nonlinear procedure monitoring. After feature normalization, suppressing environmental effects, the control charts (Hotelling T² and SPE statistics) is implemented to novelty detection and distinguish damage in structures. The hybrid PCA-KPCA technique is compared to KPCA by applying support vector machine (SVM) to evaluate the effectiveness of its performance in detecting damage. The proposed method is verified through numerical and full-scale studies (a Bridge Health Monitoring (BHM) Benchmark Problem and a cable-stayed bridge in China). The results demonstrate that the proposed method can detect the structural damage accurately and reduce false alarms by suppressing the effects and interference of environmental variations.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.5
• /
• pp.600-608
• /
• 2020

Machine learning constructs an objective function from learning data, and predicts the result of the data generated by checking the objective function through test data. In machine learning, input data is subjected to a normalisation process through a preprocessing. In the case of numerical data, normalization is standardized by using the average and standard deviation of the input data. In the case of nominal data, which is non-numerical data, it is converted into a one-hot code form. However, this preprocessing alone cannot solve the problem. For this reason, we propose a method that uses ontology to normalize input data in this paper. The test data for this uses the received signal strength indicator (RSSI) value of the Wi-Fi device collected from the mobile device. These data are solved through ontology because they includes noise and heterogeneous problems.

• Knowledge Management Research
• /
• v.22 no.4
• /
• pp.27-40
• /
• 2021

Click-Through Rate (CTR) prediction is a key function that determines the ranking of candidate items in the recommendation system and recommends high-ranking items to reduce customer information overload and achieve profit maximization through sales promotion. The fields of natural language processing and image classification are achieving remarkable growth through the use of deep neural networks. Recently, a transformer model based on an attention mechanism, differentiated from the mainstream models in the fields of natural language processing and image classification, has been proposed to achieve state-of-the-art in this field. In this study, we present a method for improving the performance of a transformer model for CTR prediction. In order to analyze the effect of discrete and categorical CTR data characteristics different from natural language and image data on performance, experiments on embedding regularization and transformer normalization are performed. According to the experimental results, it was confirmed that the prediction performance of the transformer was significantly improved when the L2 generalization was applied in the embedding process for CTR data input processing and when batch normalization was applied instead of layer normalization, which is the default regularization method, to the transformer model.

• Korean Journal of Fisheries and Aquatic Sciences
• /
• v.51 no.1
• /
• pp.95-106
• /
• 2018

To evaluate appropriate reference genes for the normalization of quantitative reverse transcription PCR (RT-qPCR) data with embryonic and larval samples from Russian sturgeon Acipenser gueldenstaedtii, the expression stability of eight candidate housekeeping genes, including beta-actin (ACTB), elongation factor-1A (EF1A), glyceraldehyde-3-phosphate dehydrogenase (GAPDH), histone 2A (H2A), ribosomal protein L5 (RPL5), ribosomal protein L7 (RPL7), succinate dehydrogenase (SDHA), and ubiquitin-conjugating enzyme E2 (UBE2A), were tested using embryonic samples from 12 developmental stages and larval samples from 11 ontogenic stages. Based on the stability rankings from three statistic software packages, geNorm, NormFinder, and BestKeeper, the expression stability of the embryonic subset was ranked as UBE2A>H2A>SDHA>GAPDH>RPL5>EF1A>ACTB>RPL7. On the other hand, the ranking in the larval subset was determined as UBE2A>GAPDH>SDHA>RPL5>RPL7>H2A>EF1A>AC TB. When the two subsets were combined, the overall ranking was UBE2A>SDHA>H2A>RPL5>GAPDH>EF1A>ACTB>RPL7. Taken together, our data suggest that UBE2A and SDHA are recommended as suitable references for developmental and ontogenic samples of this sturgeon species, whereas traditional housekeepers such as ACTB and GAPDH may not be suitable candidates.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.19 no.6
• /
• pp.773-778
• /
• 2009

This paper studied the training methods less affected by the emotional variation for the development of the robust speech recognition system. For this purpose, the effect of emotional variations on the speech signal were studied using speech database containing various emotions. The performance of the speech recognition system trained by using the speech signal containing no emotion is deteriorated if the test speech signal contains the emotions because of the emotional difference between the test and training data. In this study, it is observed that vocal tract length of the speaker is affected by the emotional variation and this effect is one of the reasons that makes the performance of the speech recognition system worse. In this paper, vocal tract normalization method is used to develop the robust speech recognition system for emotional variations. Experimental results from the isolated word recognition using HMM showed that the vocal tract normalization method reduced the error rate of the conventional recognition system by 41.9% when emotional test data was used.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.7
• /
• pp.2390-2406
• /
• 2022

Scene text recognition has important application value and attracted the interest of plenty of researchers. At present, many methods have achieved good results, but most of the existing approaches attempt to improve the performance of scene text recognition from the image level. They have a good effect on reading regular scene texts. However, there are still many obstacles to recognizing text on low-quality images such as curved, occlusion, and blur. This exacerbates the difficulty of feature extraction because the image quality is uneven. In addition, the results of model testing are highly dependent on training data, so there is still room for improvement in scene text recognition methods. In this work, we present a natural scene text recognizer to improve the recognition performance from the feature level, which contains feature representation and feature enhancement. In terms of feature representation, we propose an efficient feature extractor combined with Representative Batch Normalization and ResNet. It reduces the dependence of the model on training data and improves the feature representation ability of different instances. In terms of feature enhancement, we use a feature enhancement network to expand the receptive field of feature maps, so that feature maps contain rich feature information. Enhanced feature representation capability helps to improve the recognition performance of the model. We conducted experiments on 7 benchmarks, which shows that this method is highly competitive in recognizing both regular and irregular texts. The method achieved top1 recognition accuracy on four benchmarks of IC03, IC13, IC15, and SVTP.