• The KIPS Transactions:PartB
• /
• v.12B no.1 s.97
• /
• pp.47-56
• /
• 2005

A moving object has a various features that its spatial location, shape, and size are changed as time goes. In addition, the moving object has both temporal feature and spatial feature. It is one of the highly interested feature information in video data. In this paper, we propose an efficient content-based multimedia information retrieval system, so tailed ECoMOT which enables user to retrieve video data by using a trajectory information of moving objects in video data. The ECoMOT includes several novel techniques to achieve content-based retrieval using moving objects' trajectories : (1) Muitiple trajectory modeling technique to model the multiple trajectories composed of several moving objects; (2) Multiple similar trajectory retrieval technique to retrieve more similar trajectories by measuring similarity between a given two trajectories composed of several moving objects; (3) Superimposed signature-based trajectory indexing technique to effectively search corresponding trajectories from a large trajectory databases; (4) convenient trajectory extraction, query generation, and retrieval interface based on graphic user interface

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.43 no.1 s.307
• /
• pp.53-66
• /
• 2006

This paper proposes the combined image retrieval system that gives the same relevance as exhaustive search method while its performance can be considerably improved. This system is combined with two different retrieval methods and each gives the same results that full exhaustive search method does. Both of them are two-stage method. One uses condensation of feature vectors, and the other uses binary-tree clustering. These two methods extract the candidate images that always include correct answers at the first stage, and then filter out the incorrect images at the second stage. Inasmuch as these methods use equal algorithm, they can get the same result as full exhaustive search. The first method condenses the dimension of feature vectors, and it uses these condensed feature vectors to compute similarity of query and images in database. It can be found that there is an optimal condensation ratio which minimizes the overall retrieval time. The optimal ratio is applied to first stage of this method. Binary-tree clustering method, searching with recursive 2-means clustering, classifies each cluster dynamically with the same radius. For preserving relevance, its range of query has to be compensated at first stage. After candidate clusters were selected, final results are retrieved by computing similarities again at second stage. The proposed method is combined with above two methods. Because they are not dependent on each other, combined retrieval system can make a remarkable progress in performance.

• Journal of the Korea Computer Graphics Society
• /
• v.24 no.4
• /
• pp.29-38
• /
• 2018

In this paper, we propose an automatic segmentation method of meniscus in knee MR images by automatic meniscus localization, multi-atlas-based locally-weighted voting, and patch-based edge feature classification. First, after segmenting the bone and knee articular cartilage, the volume of interest of the meniscus is automatically localized. Second, the meniscus is segmented by multi-atlas-based locally-weighted voting taking into account the weights of shape and intensity distribution in the volume of interest of the meniscus. Finally, to remove leakage to the collateral ligaments with similar intensity, meniscus is refined using patch-based edge feature classification considering shape and distance weights. Dice similarity coefficient between proposed method and manual segmentation were 80.13% of medial meniscus and 80.81 % for lateral meniscus, and showed better results of 7.25% for medial meniscus and 1.31% for lateral meniscus compared to the multi-atlas-based locally-weighted voting.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.17 no.2
• /
• pp.15-23
• /
• 2017

The person Re-identification is the most challenging part of computer vision due to the significant changes in human pose and background clutter with occlusions. The picture from non-overlapping cameras enhance the difficulty to distinguish some person from the other. To reach a better performance match, most methods use feature selection and distance metrics separately to get discriminative representations and proper distance to describe the similarity between person and kind of ignoring some significant features. This situation has encouraged us to consider a novel method to deal with this problem. In this paper, we proposed an enhanced recurrent neural network with three-tier hierarchical network for person re-identification. Specifically, the proposed recurrent neural network (RNN) model contain an iterative expectation maximum (EM) algorithm and three-tier Hierarchical network to jointly learn both the discriminative features and metrics distance. The iterative EM algorithm can fully use of the feature extraction ability of convolutional neural network (CNN) which is in series before the RNN. By unsupervised learning, the EM framework can change the labels of the patches and train larger datasets. Through the three-tier hierarchical network, the convolutional neural network, recurrent network and pooling layer can jointly be a feature extractor to better train the network. The experimental result shows that comparing with other researchers' approaches in this field, this method also can get a competitive accuracy. The influence of different component of this method will be analyzed and evaluated in the future research.

• KIPS Transactions on Computer and Communication Systems
• /
• v.11 no.10
• /
• pp.363-372
• /
• 2022

Since the recent COVID-19 Pandemic, the ransomware fandom has intensified along with the expansion of remote work. Currently, anti-virus vaccine companies are trying to respond to ransomware, but traditional file signature-based static analysis can be neutralized in the face of diversification, obfuscation, variants, or the emergence of new ransomware. Various studies are being conducted for such ransomware detection, and detection studies using signature-based static analysis and behavior-based dynamic analysis can be seen as the main research type at present. In this paper, the frequency of ".text Section" Opcode and the Native API used in practice was extracted, and the association between feature information selected using K-means Clustering algorithm, Cosine Similarity, and Pearson correlation coefficient was analyzed. In addition, Through experiments to classify and detect worms among other malware types and Cerber-type ransomware, it was verified that the selected feature information was specialized in detecting specific ransomware (Cerber). As a result of combining the finally selected feature information through the above verification and applying it to machine learning and performing hyper parameter optimization, the detection rate was up to 93.3%.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.45 no.1
• /
• pp.25-36
• /
• 2008

Eye localization means localization of the center of the pupils, and is necessary for face recognition and related applications. Most of eye localization methods reported so far still need to be improved about robustness as well as precision for successful applications. In this paper, we propose a robust eye localization method using multi-scale Gabor feature vectors without big computational burden. The eye localization method using Gabor feature vectors is already employed in fuck as EBGM, but the method employed in EBGM is known not to be robust with respect to initial values, illumination, and pose, and may need extensive search range for achieving the required performance, which may cause big computational burden. The proposed method utilizes multi-scale approach. The proposed method first tries to localize eyes in the lower resolution face image by utilizing Gabor Jet similarity between Gabor feature vector at an estimated initial eye coordinates and the Gabor feature vectors in the eye model of the corresponding scale. Then the method localizes eyes in the next scale resolution face image in the same way but with initial eye points estimated from the eye coordinates localized in the lower resolution images. After repeating this process in the same way recursively, the proposed method funally localizes eyes in the original resolution face image. Also, the proposed method provides an effective illumination normalization to make the proposed multi-scale approach more robust to illumination, and additionally applies the illumination normalization technique in the preprocessing stage of the multi-scale approach so that the proposed method enhances the eye detection success rate. Experiment results verify that the proposed eye localization method improves the precision rate without causing big computational overhead compared to other eye localization methods reported in the previous researches and is robust to the variation of post: and illumination.

• Journal of Intelligence and Information Systems
• /
• v.25 no.2
• /
• pp.141-166
• /
• 2019

Recently, channels like social media and SNS create enormous amount of data. In all kinds of data, portions of unstructured data which represented as text data has increased geometrically. But there are some difficulties to check all text data, so it is important to access those data rapidly and grasp key points of text. Due to needs of efficient understanding, many studies about text summarization for handling and using tremendous amounts of text data have been proposed. Especially, a lot of summarization methods using machine learning and artificial intelligence algorithms have been proposed lately to generate summary objectively and effectively which called "automatic summarization". However almost text summarization methods proposed up to date construct summary focused on frequency of contents in original documents. Those summaries have a limitation for contain small-weight subjects that mentioned less in original text. If summaries include contents with only major subject, bias occurs and it causes loss of information so that it is hard to ascertain every subject documents have. To avoid those bias, it is possible to summarize in point of balance between topics document have so all subject in document can be ascertained, but still unbalance of distribution between those subjects remains. To retain balance of subjects in summary, it is necessary to consider proportion of every subject documents originally have and also allocate the portion of subjects equally so that even sentences of minor subjects can be included in summary sufficiently. In this study, we propose "subject-balanced" text summarization method that procure balance between all subjects and minimize omission of low-frequency subjects. For subject-balanced summary, we use two concept of summary evaluation metrics "completeness" and "succinctness". Completeness is the feature that summary should include contents of original documents fully and succinctness means summary has minimum duplication with contents in itself. Proposed method has 3-phases for summarization. First phase is constructing subject term dictionaries. Topic modeling is used for calculating topic-term weight which indicates degrees that each terms are related to each topic. From derived weight, it is possible to figure out highly related terms for every topic and subjects of documents can be found from various topic composed similar meaning terms. And then, few terms are selected which represent subject well. In this method, it is called "seed terms". However, those terms are too small to explain each subject enough, so sufficient similar terms with seed terms are needed for well-constructed subject dictionary. Word2Vec is used for word expansion, finds similar terms with seed terms. Word vectors are created after Word2Vec modeling, and from those vectors, similarity between all terms can be derived by using cosine-similarity. Higher cosine similarity between two terms calculated, higher relationship between two terms defined. So terms that have high similarity values with seed terms for each subjects are selected and filtering those expanded terms subject dictionary is finally constructed. Next phase is allocating subjects to every sentences which original documents have. To grasp contents of all sentences first, frequency analysis is conducted with specific terms that subject dictionaries compose. TF-IDF weight of each subjects are calculated after frequency analysis, and it is possible to figure out how much sentences are explaining about each subjects. However, TF-IDF weight has limitation that the weight can be increased infinitely, so by normalizing TF-IDF weights for every subject sentences have, all values are changed to 0 to 1 values. Then allocating subject for every sentences with maximum TF-IDF weight between all subjects, sentence group are constructed for each subjects finally. Last phase is summary generation parts. Sen2Vec is used to figure out similarity between subject-sentences, and similarity matrix can be formed. By repetitive sentences selecting, it is possible to generate summary that include contents of original documents fully and minimize duplication in summary itself. For evaluation of proposed method, 50,000 reviews of TripAdvisor are used for constructing subject dictionaries and 23,087 reviews are used for generating summary. Also comparison between proposed method summary and frequency-based summary is performed and as a result, it is verified that summary from proposed method can retain balance of all subject more which documents originally have.

• Journal of the Korea Computer Graphics Society
• /
• v.24 no.3
• /
• pp.73-81
• /
• 2018

In this paper, we present a motion-driven rowing simulation framework that allows multiple players to row a boat together by their harmonized movements. In the actual rowing game, it is crucial for the players to synchronize their rowing with respect to time and pose so as to accelerate the boat. Inspired by this interesting feature, we measure the motion similarity among multiple players in real time while they are doing rowing motions and use it to control the velocity of the boat in a virtual environment. We also employ game components such as catching an item which can accelerate or decelerate the boat depending on its type for a moment once it has been obtained by synchronized catching behaviors of the players. By these components, the players can be encouraged to more actively participate in the training for a good teamwork to produce harmonized rowing movements Our methods for the motion recognition for rowing and item catch require the tracking data only for the head and the both hands and are fast enough to facilitate the real-time performance. In order to enhance immersiveness of the virtual environment, we project the rowing simulation result on a wide curved screen.

• The KIPS Transactions:PartD
• /
• v.13D no.4 s.107
• /
• pp.455-462
• /
• 2006

Recently, the filtering approach using vector approximation such as VA-file[1] or LPC-file[2] have been proposed to support similarity search in high dimensional data space. This approach filters out many irrelevant vectors by calculating the approximate distance from a query vector using the compact approximations of vectors in database. Accordingly, the total elapsed time for similarity search is reduced because the disk I/O time is eliminated by reading the compact approximations instead of original vectors. However, the search time of the VA-file or LPC-file is not much lessened compared to the brute-force search because it requires a lot of computations for calculating the approximate distance. This paper proposes a new bitmap index structure in order to minimize the calculating time. To improve the calculating speed, a specific value of an object is saved in a bit pattern that shows a spatial position of the feature vector on a data space, and the calculation for a distance between objects is performed by the XOR bit calculation that is much faster than the real vector calculation. According to the experiment, the method that this paper suggests has shortened the total searching time to the extent of about one fourth of the sequential searching time, and to the utmost two times of the existing methods by shortening the great deal of calculating time, although this method has a longer data reading time compared to the existing vector approximation based approach. Consequently, it can be confirmed that we can improve even more the searching performance by shortening the calculating time for filtering of the existing vector approximation methods when the database speed is fast enough.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.5
• /
• pp.461-467
• /
• 2020

In noisy and multi-speaker environments, the performance of speech recognition is unavoidably lower than in a clean environment. To improve speech recognition, in this paper, the signal of the speaker of interest is extracted from the mixed speech signals with multiple speakers. The VoiceFilter model is used to effectively separate overlapped speech signals. In this work, clustering by Probabilistic Linear Discriminant Analysis (PLDA) similarity score was employed to detect the speech signal of the interested speaker, which is used as the reference speaker to VoiceFilter-based separation. Therefore, by utilizing the speaker feature extracted from the detected speech by the proposed clustering method, this paper propose a speaker diarization system using only the mixed speech without an explicit reference speaker signal. We use phone-dataset consisting of two speakers to evaluate the performance of the speaker diarization system. Source to Distortion Ratio (SDR) of the operator (Rx) speech and customer speech (Tx) are 5.22 dB and -5.22 dB respectively before separation, and the results of the proposed separation system show 11.26 dB and 8.53 dB respectively.