• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.6 no.5
• /
• pp.777-782
• /
• 2002

This study will offer multimodal recognition instead of an existing monomodal bioinfomatics by using facial multi-feature to improve the accuracy of recognition and to consider the convenience of user . Each bioinfomatics vector can be found by the following ways. For a face, the feature is calculated by principal component analysis with wavelet multiresolution. For a lip, a filter is used to find out an equation to calculate the edges of the lips first. Then by using a thinning image and least square method, an equation factor can be drawn. A feature found out the facial parameter distance ratio. We've sorted backpropagation neural network and experimented with the inputs used above. Based on the experimental results we discuss the advantage and efficiency.

• Journal of the Korean Society for Precision Engineering
• /
• v.33 no.9
• /
• pp.715-721
• /
• 2016

In this paper, the classification rate of micro-cracks in silicon wafers was improved using a SVM. In case I, we investigated how feature data of micro-cracks and SVM parameters affect a classification rate. As a result, weighting vector and bias did not affect the classification rate, which was improved in case of high cost and sigmoid kernel function. Case II was performed using a more high quality image than that in case I. It was identified that learning data and input data had a large effect on the classification rate. Finally, images from cases I and II and another illumination system were used in case III. In spite of different condition images, good classification rates was achieved. Critical points for micro-crack classification improvement are SVM parameters, kernel function, clustered feature data, and experimental conditions. In the future, excellent results could be obtained through SVM parameter tuning and clustered feature data.

• Journal of Internet Computing and Services
• /
• v.22 no.3
• /
• pp.9-16
• /
• 2021

This study aims to develop a human activity recognition (HAR) system as a Deep-Learning (DL) classification model, distinguishing various human activities. We solely rely on the signals from a wristband accelerometer worn by a person for the user's convenience. 3-axis sequential acceleration signal data are gathered within a predefined time-window-slice, and they are used as input to the classification system. We are particularly interested in developing a Deep-Learning model that can outperform conventional machine learning classification performance. A total of 13 activities based on the laboratory experiments' data are used for the initial performance comparison. We have improved classification performance using the Convolutional Neural Network (CNN) combined with an auto-encoder feature reduction and parameter tuning. With various publically available HAR datasets, we could also achieve significant improvement in HAR classification. Our CNN model is also compared against Recurrent-Neural-Network(RNN) with Long Short-Term Memory(LSTM) to demonstrate its superiority. Noticeably, our model could distinguish both general activities and near-identical activities such as sitting down on the chair and floor, with almost perfect classification accuracy.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.9
• /
• pp.2458-2482
• /
• 2023

Medical image segmentation techniques based on convolution neural networks indulge in feature extraction triggering redundancy of parameters and unsatisfactory target localization, which outcomes in less accurate segmentation results to assist doctors in diagnosis. In this paper, we propose a multi-level semantic-rich encoding-decoding network, which consists of a Pooling-Conv-Former (PCFormer) module and a Cbam-Dilated-Transformer (CDT) module. In the PCFormer module, it is used to tackle the issue of parameter explosion in the conservative transformer and to compensate for the feature loss in the down-sampling process. In the CDT module, the Cbam attention module is adopted to highlight the feature regions by blending the intersection of attention mechanisms implicitly, and the Dilated convolution-Concat (DCC) module is designed as a parallel concatenation of multiple atrous convolution blocks to display the expanded perceptual field explicitly. In addition, MultiHead Attention-DwConv-Transformer (MDTransformer) module is utilized to evidently distinguish the target region from the background region. Extensive experiments on medical image segmentation from Glas, SIIM-ACR, ISIC and LGG demonstrated that our proposed network outperforms existing advanced methods in terms of both objective evaluation and subjective visual performance.

• Journal of Broadcast Engineering
• /
• v.23 no.1
• /
• pp.86-92
• /
• 2018

In this paper, we propose a method for coding history detection of digital speech signal. In digital speech communication and storage, the signal is encoded to reduce the number of bits. Therefore, when a speech signal waveform is given, we need to detect its coding history so that we can determine whether the signal is an original or an coded one, and if coded, determine the number of times of coding. In this paper, we propose a coding history detection method for 12.2kbps AMR codec in terms of original, single coding, and double coding. The proposed method extracts a speech-specific feature vector from the given speech, and models the feature vector using a deep neural network. We confirm that the proposed feature vector provides better performance in coding history detection than the feature vector computed from the general spectrogram.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.8 no.4
• /
• pp.561-567
• /
• 2013

Some of the feature vectors for speech recognition are rare and unusual. These patterns lead to overfitting for the parameters of the speech recognition system and, as a result, cause structural risks in the system that hinder the good performance in recognition. In this paper, as a method of removing these unusual patterns, we try to exclude vectors whose norms are larger than a specified cutoff value and then train the speech recognition system. The objective of this study is to exclude as many unusual feature vectors under the condition of no significant degradation in the speech recognition error rate. For this purpose, we introduce a cutoff parameter and investigate the resultant effect on the speaker-independent speech recognition of isolated words by using FVQ(Fuzzy Vector Quantization)/HMM(Hidden Markov Model). Experimental results showed that roughly 3%~6% of the feature vectors might be considered as unusual, and therefore be excluded without deteriorating the speech recognition accuracy.

• Journal of Korea Multimedia Society
• /
• v.14 no.6
• /
• pp.742-751
• /
• 2011

This paper presents a robust 3D model hashing dependent on key and parameter by using heat kernel signature (HKS), which is special shape feature descriptor, In the proposed hashing, we calculate HKS coefficients of local and global time scales from eigenvalue and eigenvector of Mesh Laplace operator and cluster pairs of HKS coefficients to 2D square cells and calculate feature coefficients by the distance weights of pairs of HKS coefficients on each cell. Then we generate the binary hash through binarizing the intermediate hash that is the combination of the feature coefficients and the random coefficients. In our experiment, we evaluated the robustness against geometrical and topological attacks and the uniqueness of key and model and also evaluated the model space by estimating the attack intensity that can authenticate 3D model. Experimental results verified that the proposed scheme has more the improved performance than the conventional hashing on the robustness, uniqueness, model space.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.38 no.4
• /
• pp.381-391
• /
• 2001

This paper presents an efficient approach to build an image mosaics from image sequences. Unlike general panoramic stitching methods, which usually require some geometrical feature points or solve the iterative nonlinear equations, our algorithm can directly recover the 8-parameter planar perspective transforms. We use four quasi-feature points in order to compute the projective transform between two images. This feature is based on the graylevel distribution and defined in the overlap area between two images. Therefore the proposed algorithm can reduce the total amount of the computation. We also present an algorithm lot efficiently matching the correspondence of the extracted feature. The proposed algorithm is applied to various images to estimate its performance and. the simulation results present that our algorithm can find the correct correspondence and build an image mosaics.

• Journal of Biosystems Engineering
• /
• v.28 no.1
• /
• pp.65-76
• /
• 2003

The analysis of cow body parameters is important to provide some useful information fur cow management and cow evaluation. Present methods give many stresses to cows because they are invasive and constrain cow postures during measurement of body parameters. This study was conducted to develop the stereo vision system fur non-invasive analysis of cow body features. Body feature parameters of 16 heads at two farms(A, B) were measured using scales and nineteen stereo images of them with walking postures were captured under outdoor illumination. In this study, the camera calibration and inverse perspective transformation technique was established fer the stereo vision system. Two calibration results were presented for farm A and fm B, respectively because setup distances from camera to cow were 510 cm at farm A and 630cm at farm B. Calibration error values fer the stereo vision system were within 2 cm for farm A and less than 4.9 cm for farm B. Eleven feature points of cow body were extracted on stereo images interactively and five assistant points were determined by computer program. 3D world coordinates for these 15 points were calculated by computer program and also used for calculation of cow body parameters such as withers height. pelvic arch height. body length. slope body length. chest depth and chest width. Measured errors for body parameters were less than 10% for most cows. For a few cow. measured errors for slope body length and chest width were more than 10% due to searching errors fer their feature points at inside-body positions. Equation for chest girth estimated by chest depth and chest width was presented. Maximum of estimated error fur chest girth was within 10% of real values and mean value of estimated error was 8.2cm. The analysis of cow body parameters using stereo vision system were successful although body shape on the binocular stereo image was distorted due to cow movements.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.35 no.11C
• /
• pp.929-939
• /
• 2010

The growth of multimedia technology forces the development of video detection for large database management and illegal copy detection. To meet this demand, this paper proposes a fast video detection method to apply to a large database. The fast video detection algorithm uses spatial features using the gray value distribution from frames and temporal features using the temporal similarity map. We form the video signature using the extracted spatial feature and temporal feature, and carry out a stepwise matching method. The performance was evaluated by accuracy, extraction and matching time, and signature size using the original videos and their modified versions such as brightness change, lossy compression, text/logo overlay. We show empirical parameter selection and the experimental results for the simple matching method using only spatial feature and compare the results with existing algorithms. According to the experimental results, the proposed method has good performance in accuracy, processing time, and signature size. Therefore, the proposed fast detection algorithm is suitable for video detection with the large database.