• Journal of Korea Multimedia Society
• /
• v.13 no.2
• /
• pp.225-236
• /
• 2010

This study proposes a method for the recognition of emergency situation according to the bimodal information of camera image sensor and gravity sensor. This method can recognize emergency condition by mutual cooperation and compensation between sensors even when one of the sensors malfunction, the user does not carry gravity sensor, or in the place like bathroom where it is hard to acquire camera images. This paper implemented HMM(Hidden Markov Model) based learning and recognition algorithm to recognize actions such as walking, sitting on floor, sitting at sofa, lying and fainting motions. Recognition rate was enhanced when image feature vectors and gravity feature vectors are combined in learning and recognition process. Also, this method maintains high recognition rate by detecting moving object through adaptive background model even in various illumination changes.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.52 no.1
• /
• pp.162-172
• /
• 2015

In this paper, we propose an adaptive detection algorithm of footstep and a classification algorithm for activities of the detected footstep. The proposed algorithm can detect and classify whole movement as well as individual and irregular activities, since it does not use continuous footstep signals which are used by most previous research. For classifying movement, we use feature vectors obtained from frequency spectrum from FFT, CWT, AR model and image of AR spectrogram. With SVM classifier, we obtain classification accuracy of single footstep activities over 90% when feature vectors using AR spectrogram image are used.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.18 no.1
• /
• pp.32-38
• /
• 1993

In this paper a effective classification method for character recognition is proposed. The existing classification methods select candidates by comparing an unknown input character, with all the standard patterns based on the similarity measur. The proposed method, however, groups similiar characters together and uses their average distance as representative value of the group. We divided the character region into several sub-region and applied ISODATA algorithm to partial vectors of each sub-region to anstruct appropriate number of groups. After computing the distance between partial feature vector and its mapping group, we could collect all the information of input character ultimately. The proposed method showed improvement in the processing speed and certainty in classification than the existing methods.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.43 no.1 s.307
• /
• pp.9-16
• /
• 2006

RDA(Relational Discriminant Analysis) is a way of finding classifiers based on the dissimilarity measures among the prototypes extracted from feature vectors instead of the feature vectors themselves. Therefore, the accuracy of the RDA classifier is dependent on the methods of selecting prototypes and measuring proximities. In this paper we propose to utilize PRS(Prototype Reduction Schemes) and Mahalanobis distances to devise a method of increasing classification accuracies. Our experimental results demonstrate that the proposed mechanism increases the classification accuracy compared with the conventional approaches for samples involving real-life data sets as well as artificial data sets.

• Journal of rehabilitation welfare engineering & assistive technology
• /
• v.9 no.4
• /
• pp.285-291
• /
• 2015

In this paper, we propose a classification algorithm for postures of sitting person by using IMU(inertial measurement unit). This algorithm uses PCA(principle component analysis) for decreasing the number of feature vectors to three and SVM(support vector machine) with RBF(radial basis function) kernel for classifying posture types. In order to collect the data, we designed neckband-shaped earphones with IMU, and applied it to three subjects who are healthy adults. Subjects were experimented three sitting postures, which are neutral posture, smartphoning, and writing. As the result, our PCA-SVM algorithm showed 95% confidence while the dimension of the feature vectors was reduced to 25%.

• Journal of Internet Computing and Services
• /
• v.5 no.3
• /
• pp.57-62
• /
• 2004

In this paper, an efficient speaker identification based on robust vector quantizationprincipal component analysis (VQ-PCA) is proposed to solve the problems from outliers and high dimensionality of training feature vectors in speaker identification, Firstly, the proposed method partitions the data space into several disjoint regions by roust VQ based on M-estimation. Secondly, the robust PCA is obtained from the covariance matrix in each region. Finally, our method obtains the Gaussian Mixture model (GMM) for speaker from the transformed feature vectors with reduced dimension by the robust PCA in each region, Compared to the conventional GMM with diagonal covariance matrix, under the same performance, the proposed method gives faster results with less storage and, moreover, shows robust performance to outliers.

• Journal of Korea Multimedia Society
• /
• v.6 no.5
• /
• pp.753-768
• /
• 2003

In this paper, we propose a multi-dimensional index structure, tailed a VA(Vector Approximate)-tree that is constructed with vector approximates of multi-dimensional feature vectors. To save storage space for index structures, the VA-tree employs vector approximation concepts of VA-file that presents feature vectors with much smaller number of bits than original value. Since the VA-tree is a tree structure, it does not suffer from performance degradation owing to the increase of data. Also, even though the VA-tree is MBR(Minimum Bounding Region) based tree structure like a R-tree, its split algorithm never allows overlap between MBRs. We show through various experiments that our proposed VA-tree is a suitable index structure for large amount of multi-dimensional data.

• The Journal of the Acoustical Society of Korea
• /
• v.13 no.1E
• /
• pp.71-81
• /
• 1994

In this paper, we propose a pseudo-cepstral representation of line spectrum pair(LSP) frequencies and evaluate speech recognition performance with cepstral lift using the pseudo-cepstrum. The pseudo-cepstrum corresponding to LSP frequencies is derived by approxmating the relationship between LPC-cepstrum and LSP frequencies. Three cepstral liftering procedures are applied to the pseudo-cepstrum to improve the performance of speech recognition. They are the root-power-sums ligter, the general exponential lifter, and the bandpass lifter. Then, the liftered psedudo-cepstra are warped into a mel-frequency scale to obtain feature vectors for speech recognition. Among the three lifters, the general exponential lifter results in the best performance on speech recognition. When we use the proposed pseudo-cepstra feature vectors for recognizing noisy speech, the signal-to-noise ratio (SNR) improvement of about 5~10dB LSP is obtained.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.9
• /
• pp.4567-4587
• /
• 2017

With the growth of the Internet and the extensive applications of image editing software, it has become easier to manipulate digital images without leaving obvious traces. Copy-move is one of the most common techniques for image forgery. Image blind forensics is an effective technique for detecting tampered images. This paper proposes an improved copy-move forgery detection method based on the discrete cosine transform (DCT). The quantized DCT coefficients, which are feature representations of image blocks, are truncated using a truncation factor to reduce the feature dimensions. A method for judging whether two image blocks are similar is proposed to improve the accuracy of similarity judgments. The main transfer vectors whose frequencies exceed a threshold are found to locate the copied and pasted regions in forged images. Several experiments are conducted to test the practicability of the proposed algorithm using images from copy-move databases and to evaluate its robustness against post-processing methods such as additive white Gaussian noise (AWGN), Gaussian blurring, and JPEG compression. The results of experiments show that the proposed scheme effectively detects both copied region and pasted region of forged images and that it is robust to the post-processing methods mentioned above.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.16 no.9
• /
• pp.1878-1883
• /
• 2012

Generally, in the underwater target recognition, feature vectors are extracted from the target signal utilizing spatial information according to target shape/material characteristics. And, various signal processing techniques have been studied to extract feature vectors which is less sensitive to the location of the receiver. In this paper, we analyzed the characteristics of synthesized underwater objects using canonical correlation analysis method which is relatively less sensitive to the location of receiver. Canonical correlation analysis is applied to two consecutive backscattered sonar returns at different aspect angles to analyze the correlation characteristics in multi-aspect environment.