• Journal of Intelligence and Information Systems
• /
• v.21 no.2
• /
• pp.131-150
• /
• 2015

Omission of noun phrases for obligatory cases is a common phenomenon in sentences of Korean and Japanese, which is not observed in English. When an argument of a predicate can be filled with a noun phrase co-referential with the title, the argument is more easily omitted in Encyclopedia texts. The omitted noun phrase is called a zero anaphor or zero pronoun. Encyclopedias like Wikipedia are major source for information extraction by intelligent application systems such as information retrieval and question answering systems. However, omission of noun phrases makes the quality of information extraction poor. This paper deals with the problem of developing a system that can restore omitted noun phrases in encyclopedia documents. The problem that our system deals with is almost similar to zero anaphora resolution which is one of the important problems in natural language processing. A noun phrase existing in the text that can be used for restoration is called an antecedent. An antecedent must be co-referential with the zero anaphor. While the candidates for the antecedent are only noun phrases in the same text in case of zero anaphora resolution, the title is also a candidate in our problem. In our system, the first stage is in charge of detecting the zero anaphor. In the second stage, antecedent search is carried out by considering the candidates. If antecedent search fails, an attempt made, in the third stage, to use the title as the antecedent. The main characteristic of our system is to make use of a structural SVM for finding the antecedent. The noun phrases in the text that appear before the position of zero anaphor comprise the search space. The main technique used in the methods proposed in previous research works is to perform binary classification for all the noun phrases in the search space. The noun phrase classified to be an antecedent with highest confidence is selected as the antecedent. However, we propose in this paper that antecedent search is viewed as the problem of assigning the antecedent indicator labels to a sequence of noun phrases. In other words, sequence labeling is employed in antecedent search in the text. We are the first to suggest this idea. To perform sequence labeling, we suggest to use a structural SVM which receives a sequence of noun phrases as input and returns the sequence of labels as output. An output label takes one of two values: one indicating that the corresponding noun phrase is the antecedent and the other indicating that it is not. The structural SVM we used is based on the modified Pegasos algorithm which exploits a subgradient descent methodology used for optimization problems. To train and test our system we selected a set of Wikipedia texts and constructed the annotated corpus in which gold-standard answers are provided such as zero anaphors and their possible antecedents. Training examples are prepared using the annotated corpus and used to train the SVMs and test the system. For zero anaphor detection, sentences are parsed by a syntactic analyzer and subject or object cases omitted are identified. Thus performance of our system is dependent on that of the syntactic analyzer, which is a limitation of our system. When an antecedent is not found in the text, our system tries to use the title to restore the zero anaphor. This is based on binary classification using the regular SVM. The experiment showed that our system's performance is F1 = 68.58%. This means that state-of-the-art system can be developed with our technique. It is expected that future work that enables the system to utilize semantic information can lead to a significant performance improvement.

• Journal of Intelligence and Information Systems
• /
• v.17 no.3
• /
• pp.203-219
• /
• 2011

Previously common users just want to watch the video contents without any specific requirements or purposes. However, in today's life while watching video user attempts to know and discover more about things that appear on the video. Therefore, the requirements for finding multimedia or browsing information of objects that users want, are spreading with the increasing use of multimedia such as videos which are not only available on the internet-capable devices such as computers but also on smart TV and smart phone. In order to meet the users. requirements, labor-intensive annotation of objects in video contents is inevitable. For this reason, many researchers have actively studied about methods of annotating the object that appear on the video. In keyword-based annotation related information of the object that appeared on the video content is immediately added and annotation data including all related information about the object must be individually managed. Users will have to directly input all related information to the object. Consequently, when a user browses for information that related to the object, user can only find and get limited resources that solely exists in annotated data. Also, in order to place annotation for objects user's huge workload is required. To cope with reducing user's workload and to minimize the work involved in annotation, in existing object-based annotation automatic annotation is being attempted using computer vision techniques like object detection, recognition and tracking. By using such computer vision techniques a wide variety of objects that appears on the video content must be all detected and recognized. But until now it is still a problem facing some difficulties which have to deal with automated annotation. To overcome these difficulties, we propose a system which consists of two modules. The first module is the annotation module that enables many annotators to collaboratively annotate the objects in the video content in order to access the semantic data using Linked Data. Annotation data managed by annotation server is represented using ontology so that the information can easily be shared and extended. Since annotation data does not include all the relevant information of the object, existing objects in Linked Data and objects that appear in the video content simply connect with each other to get all the related information of the object. In other words, annotation data which contains only URI and metadata like position, time and size are stored on the annotation sever. So when user needs other related information about the object, all of that information is retrieved from Linked Data through its relevant URI. The second module enables viewers to browse interesting information about the object using annotation data which is collaboratively generated by many users while watching video. With this system, through simple user interaction the query is automatically generated and all the related information is retrieved from Linked Data and finally all the additional information of the object is offered to the user. With this study, in the future of Semantic Web environment our proposed system is expected to establish a better video content service environment by offering users relevant information about the objects that appear on the screen of any internet-capable devices such as PC, smart TV or smart phone.

• Progress in Medical Physics
• /
• v.26 no.4
• /
• pp.258-266
• /
• 2015

The modern radiotherapy technique which delivers a large amount of dose to patients asks to confirm the positions of patients or tumors more accurately by using X-ray projection images of high-definition. However, a rapid increase in patient's exposure and image information for CT image acquisition may be additional burden on the patient. In this study, by introducing structural similarity (SSIM) index that can effectively extract the structural information of the image, we analyze the differences between daily acquired x-ray images of a patient to verify the accuracy of patient positioning. First, for simulating a moving target, the spherical computational phantoms changing the sizes and positions were created to acquire projected images. Differences between the images were automatically detected and analyzed by extracting their SSIM values. In addition, as a clinical test, differences between daily acquired x-ray images of a patient for 12 days were detected in the same way. As a result, we confirmed that the SSIM index was changed in the range of 0.85~1 (0.006~1 when a region of interest (ROI) was applied) as the sizes or positions of the phantom changed. The SSIM was more sensitive to the change of the phantom when the ROI was limited to the phantom itself. In the clinical test, the daily change of patient positions was 0.799~0.853 in SSIM values, those well described differences among images. Therefore, we expect that SSIM index can provide an objective and quantitative technique to verify the patient position using simple x-ray images, instead of time and cost intensive three-dimensional x-ray images.

• Nuclear Medicine and Molecular Imaging
• /
• v.43 no.1
• /
• pp.55-59
• /
• 2009

Purpose: Lymphoscintigraphy and sentinel node biopsy are used in detection of axillary lymph node metastasis in breast cancer patients, but standardized technique is not established. We compared the results of the injection the morning of surgery (1 day protocol) with the subareolar injection the day before surgery (2 day protocol) with the subareolar injection in patients with breast cancer having lymphoscintigraphy and sentinel node biopsy. Materials and Methods: This study included 349 patients who underwent the breast cancer operation during 2001-2004. One hundred seventy one patients (1 day protocol, 1 hour) was injected 0.8ml of Tc-99m Tin-Colloid (37 MBq) by subareolar injection on the morning of surgery. One hundred seventy eight patients (2 day protocol, 16 hour) was injected 0.8 ml of T c-99m Tin-Colloid (185 MBq) on the afternoon before surgery. Lymphoscintigraphy was performed in sitting position and sentinel node localization was performed by hand-held gamma probe during operation. Result: In the 1 day protocol, 153 cases (89.5%) of the sentinel node were localized by lymphoscintigraphy and 150 cases (87.7%) were localized by gamma probe. In the 2 day protocol, 159 cases (89.3%) were localized by lymphoscintigraphy and 154 cases (86.5%) were localized by gamma probe. There was no significant difference in localization of sentinel node between the 1 day and the 2 day protocol by lymphoscintigraphy and gamma probe (p>0.05, p>0.05). Conclusion: There was no difference the result of localization of sentinel node with subareolar injection between the 1 day and the 2 day protocol in breast cancer patients. Because the 2 day protocol allows the enough time of performing lymphoscintigraphy, it is more useful in localization of sentinel node in breast cancer patients.

• Radiation Oncology Journal
• /
• v.27 no.4
• /
• pp.194-200
• /
• 2009

Purpose: This aim of this study was to evaluate changes in gastric volume and organ position as a result of delayed gastric emptying after a subtotal gastrectomy performed as part of the treatment of stomach cancer. Materials and Methods: The medical records of 32 patients who underwent concurrent chemoradiotherapy after a subtotal gastrectomy from March 2005 to December 2008 were reviewed. Of these, 5 patients that had more than 50 cc of residual gastric food detected at computed tomography (CT) simulation, were retrospectively enrolled in this study. Gastric volume and organ location was measured from CT images obtained before radiotherapy, twice weekly. In addition, authors evaluated the change of radiation dose distribution to planning the target volume and normal organ in a constant radiation therapy plan regardless of gastric volume variation. Results: A variation in the gastric volume was observed during the radiotherapy period (64.2~340.8 cc; mean, 188.2 cc). According to the change in gastric volume, the location of the left kidney was shifted up to 0.7 - 2.2 cm (mean, 1.2 cm) in the z-axis. Under-dose to planning target volume (V43, 79.5${\pm}$10.4%) and over-dose to left kidney (V20, 34.1${\pm}$12.1%; Mean dose, 23.5${\pm}$8.3 Gy) was expected, given that gastric volume change due to delayed gastric emptying wasn't taken into account. Conclusion: This study has shown that a great change in gastric volume and left kidney location may occur during the radiation therapy period following a subtotal gastrectomy, as a result of delayed gastric emptying. Detection of patients who experienced delayed gastric emptying and the application of gastric volume variation to radiation therapy planning will be very important.

• Radiation Oncology Journal
• /
• v.25 no.2
• /
• pp.125-133
• /
• 2007

[ $\underline{Purpose}$ ]: This study was to design and fabricate a phantom for prostate cancer brachytherapy to validate a developed program applying a 3-film technique, and to compare it with the conventional 2-film technique for determining the location of an implanted seed. $\underline{Materials\;and\;Methods}$: The images were obtained from overlapped seeds by randomly placing a maximum of 63 seeds in the anterior-posterior (AP) position and at $-30^{\circ} to $30^{\circ} at $15^{\circ} intervals. Images obtained by use of the phantom were applied to the image processing procedure, and were then processed into the development program for seed localization. In this study, cases were set where one seed overlapped, where two seeds overlapped and where none of the three views resolved all seeds. The distance between the centers of each seed to the reference seed was calculated in a prescribed region. This distance determined the location of each seed in a given band. The location of the overlapped seeds was compared with that of the 2-film technique. $\underline{Results}$: With this program, the detection rate was 92.2% (at ${\pm}15^{\circ}), 94.1% (at ${\pm}30^{\circ}) and 70.6% (compared to the use of the 2-film technique). The overlaps were caused by one or more than two seeds that overlapped; the developed program can identify the location of each seed perfectly. However, for the third case the program was not able to resolve the overlap of the seeds. $\underline{Conclusion}$: This program can be used to improve treatment outcome for the brachytherapy of prostate cancer by reducing the number of errors in the process of reconstructing the locations of perfectly overlapped seeds.

• Journal of Life Science
• /
• v.13 no.3
• /
• pp.280-290
• /
• 2003

The incorporation of RAPD markers into the previous classical and RFLP genetic linkage maps will facilitate the generation of a detailed genetic map by compensating for the lack of one type of marker in the region of interest. The objective of this paper was to present features we observed when we associated RAPD map from an intraspecific cross of a Glycine max$\times$G. max, 'Essex'$\times$PI 437654 with the public RFLP map developed from an interspecific cross of G. max$\times$G. soja. Among 27 linkage groups of RAPD map, eight linkage groups contained probe/enzyme combination RFLP markers, which allowed us the incorporation of RAPD markers into the public RFLP map. Map position rearrangement was observed. In incorporating L.G.C-3 into the public RFLP linkage group a1 and a2, both pSAC3 and pA136 region, and pA170/EcoRV and pB170/HindIII region were in opposite order, respectively. And, pk400 was localized 1.8 cM from pA96-1 and 8.4 cM from pB172 in the public RFLP map, but was localized 9.9 cM from i locus and 18.9 cM from pA85 in our study. A noticeable expansion of the map distances in the intraspecific cross of Essex and PI 437654 was also observed. Map distance between probes pA890 and pK493 in L.G.C-1 was 48.6 cM, but it was only 13.3 cM in the public RFLP map. The distances from the probe pB32-2 to pA670 and from pA670 to pA668 in L.G. C-2 were 50.9 cM and 31.7 cM, but they were 35.9 cM and 13.5 cM in the public RFLP map. The detection of duplicate loci from the same probe that were mapped on the same or/and different linkage group was another feature we observed.

• Korean Journal of Remote Sensing
• /
• v.25 no.4
• /
• pp.383-389
• /
• 2009

The main objective of the research is a feasibility study on the intertidal zone using a X-band radar satellite, TerraSAR-X. The TerraSAR-X data have been acquired in the west coast of Korea where large tidal flats, Ganghwa and Yeongjong tidal flats, are developed. Investigations include: 1) waterline and backscattering characteristics of the high resolution X-band images in tidal flats; 2) polarimetric signature of halophytes (or salt marsh plants), specifically Suaeda japonica; and 3) phase and coherence of interferometric pairs. Waterlines from TerraSAR-X data satisfy the requirement of horizontal accuracy of 60 m that corresponds to 20 cm in average height difference while current other spaceborne SAR systems could not meet the requirement. HH-polarization was the best for extraction of waterline, and its geometric position is reliable due to the short wavelength and accurate orbit control of the TerraSAR-X. A halophyte or salt marsh plant, Suaeda japonica, is an indicator of local sea level change. From X-band ground radar measurements, a dual polarization of VV/VH-pol. is anticipated to be the best for detection of the plant with about 9 dB difference at 35 degree incidence angle. However, TerraSAR-X HH/TV dual polarization was turned to be more effective for salt marsh monitoring. The HH-HV value was the maximum of about 7.9 dB at 31.6 degree incidence angle, which is fairly consistent with the results of X-band ground radar measurement. The boundary of salt marsh is effectively traceable specifically by TerraSAR-X cross-polarization data. While interferometric phase is not coherent within normal tidal flat, areas of salt marsh where the landization is preceded show coherent interferometric phases regardless of seasons or tide conditions. Although TerraSAR-X interferometry may not be effective to directly measure height or changes in tidal flat surface, TanDEM-X or other future X-band SAR tandem missions within one-day interval would be useful for mapping tidal flat topography.

• Journal of IKEEE
• /
• v.25 no.3
• /
• pp.493-500
• /
• 2021

In this paper, we propose the development of deep learning structure to improve quality of polygonal containers. The deep learning structure consists of a convolution layer, a bottleneck layer, a fully connect layer, and a softmax layer. The convolution layer is a layer that obtains a feature image by performing a convolution 3x3 operation on the input image or the feature image of the previous layer with several feature filters. The bottleneck layer selects only the optimal features among the features on the feature image extracted through the convolution layer, reduces the channel to a convolution 1x1 ReLU, and performs a convolution 3x3 ReLU. The global average pooling operation performed after going through the bottleneck layer reduces the size of the feature image by selecting only the optimal features among the features of the feature image extracted through the convolution layer. The fully connect layer outputs the output data through 6 fully connect layers. The softmax layer multiplies and multiplies the value between the value of the input layer node and the target node to be calculated, and converts it into a value between 0 and 1 through an activation function. After the learning is completed, the recognition process classifies non-circular glass bottles by performing image acquisition using a camera, measuring position detection, and non-circular glass bottle classification using deep learning as in the learning process. In order to evaluate the performance of the deep learning structure to improve quality of polygonal containers, as a result of an experiment at an authorized testing institute, it was calculated to be at the same level as the world's highest level with 99% good/defective discrimination accuracy. Inspection time averaged 1.7 seconds, which was calculated within the operating time standards of production processes using non-circular machine vision systems. Therefore, the effectiveness of the performance of the deep learning structure to improve quality of polygonal containers proposed in this paper was proven.

• Journal of Intelligence and Information Systems
• /
• v.26 no.2
• /
• pp.1-25
• /
• 2020

In this paper, we suggest an application system architecture which provides accurate, fast and efficient automatic gasometer reading function. The system captures gasometer image using mobile device camera, transmits the image to a cloud server on top of private LTE network, and analyzes the image to extract character information of device ID and gas usage amount by selective optical character recognition based on deep learning technology. In general, there are many types of character in an image and optical character recognition technology extracts all character information in an image. But some applications need to ignore non-of-interest types of character and only have to focus on some specific types of characters. For an example of the application, automatic gasometer reading system only need to extract device ID and gas usage amount character information from gasometer images to send bill to users. Non-of-interest character strings, such as device type, manufacturer, manufacturing date, specification and etc., are not valuable information to the application. Thus, the application have to analyze point of interest region and specific types of characters to extract valuable information only. We adopted CNN (Convolutional Neural Network) based object detection and CRNN (Convolutional Recurrent Neural Network) technology for selective optical character recognition which only analyze point of interest region for selective character information extraction. We build up 3 neural networks for the application system. The first is a convolutional neural network which detects point of interest region of gas usage amount and device ID information character strings, the second is another convolutional neural network which transforms spatial information of point of interest region to spatial sequential feature vectors, and the third is bi-directional long short term memory network which converts spatial sequential information to character strings using time-series analysis mapping from feature vectors to character strings. In this research, point of interest character strings are device ID and gas usage amount. Device ID consists of 12 arabic character strings and gas usage amount consists of 4 ~ 5 arabic character strings. All system components are implemented in Amazon Web Service Cloud with Intel Zeon E5-2686 v4 CPU and NVidia TESLA V100 GPU. The system architecture adopts master-lave processing structure for efficient and fast parallel processing coping with about 700,000 requests per day. Mobile device captures gasometer image and transmits to master process in AWS cloud. Master process runs on Intel Zeon CPU and pushes reading request from mobile device to an input queue with FIFO (First In First Out) structure. Slave process consists of 3 types of deep neural networks which conduct character recognition process and runs on NVidia GPU module. Slave process is always polling the input queue to get recognition request. If there are some requests from master process in the input queue, slave process converts the image in the input queue to device ID character string, gas usage amount character string and position information of the strings, returns the information to output queue, and switch to idle mode to poll the input queue. Master process gets final information form the output queue and delivers the information to the mobile device. We used total 27,120 gasometer images for training, validation and testing of 3 types of deep neural network. 22,985 images were used for training and validation, 4,135 images were used for testing. We randomly splitted 22,985 images with 8:2 ratio for training and validation respectively for each training epoch. 4,135 test image were categorized into 5 types (Normal, noise, reflex, scale and slant). Normal data is clean image data, noise means image with noise signal, relfex means image with light reflection in gasometer region, scale means images with small object size due to long-distance capturing and slant means images which is not horizontally flat. Final character string recognition accuracies for device ID and gas usage amount of normal data are 0.960 and 0.864 respectively.