• 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.

• Proceedings of the Optical Society of Korea Conference
• /
• 2003.07a
• /
• pp.60-61
• /
• 2003

A new approach to reduce the computation time of genetic algorithm (GA) for making binary phase holograms is described. Synthesized holograms having diffraction efficiency of 75.8% and uniformity of 5.8% are proven in computer simulation and experimentally demonstrated. Recently, computer-generated holograms (CGHs) having high diffraction efficiency and flexibility of design have been widely developed in many applications such as optical information processing, optical computing, optical interconnection, etc. Among proposed optimization methods, GA has become popular due to its capability of reaching nearly global. However, there exits a drawback to consider when we use the genetic algorithm. It is the large amount of computation time to construct desired holograms. One of the major reasons that the GA' s operation may be time intensive results from the expense of computing the cost function that must Fourier transform the parameters encoded on the hologram into the fitness value. In trying to remedy this drawback, Artificial Neural Network (ANN) has been put forward, allowing CGHs to be created easily and quickly (1), but the quality of reconstructed images is not high enough to use in applications of high preciseness. For that, we are in attempt to find a new approach of combiningthe good properties and performance of both the GA and ANN to make CGHs of high diffraction efficiency in a short time. The optimization of CGH using the genetic algorithm is merely a process of iteration, including selection, crossover, and mutation operators [2]. It is worth noting that the evaluation of the cost function with the aim of selecting better holograms plays an important role in the implementation of the GA. However, this evaluation process wastes much time for Fourier transforming the encoded parameters on the hologram into the value to be solved. Depending on the speed of computer, this process can even last up to ten minutes. It will be more effective if instead of merely generating random holograms in the initial process, a set of approximately desired holograms is employed. By doing so, the initial population will contain less trial holograms equivalent to the reduction of the computation time of GA's. Accordingly, a hybrid algorithm that utilizes a trained neural network to initiate the GA's procedure is proposed. Consequently, the initial population contains less random holograms and is compensated by approximately desired holograms. Figure 1 is the flowchart of the hybrid algorithm in comparison with the classical GA. The procedure of synthesizing a hologram on computer is divided into two steps. First the simulation of holograms based on ANN method [1] to acquire approximately desired holograms is carried. With a teaching data set of 9 characters obtained from the classical GA, the number of layer is 3, the number of hidden node is 100, learning rate is 0.3, and momentum is 0.5, the artificial neural network trained enables us to attain the approximately desired holograms, which are fairly good agreement with what we suggested in the theory. The second step, effect of several parameters on the operation of the hybrid algorithm is investigated. In principle, the operation of the hybrid algorithm and GA are the same except the modification of the initial step. Hence, the verified results in Ref [2] of the parameters such as the probability of crossover and mutation, the tournament size, and the crossover block size are remained unchanged, beside of the reduced population size. The reconstructed image of 76.4% diffraction efficiency and 5.4% uniformity is achieved when the population size is 30, the iteration number is 2000, the probability of crossover is 0.75, and the probability of mutation is 0.001. A comparison between the hybrid algorithm and GA in term of diffraction efficiency and computation time is also evaluated as shown in Fig. 2. With a 66.7% reduction in computation time and a 2% increase in diffraction efficiency compared to the GA method, the hybrid algorithm demonstrates its efficient performance. In the optical experiment, the phase holograms were displayed on a programmable phase modulator (model XGA). Figures 3 are pictures of diffracted patterns of the letter "0" from the holograms generated using the hybrid algorithm. Diffraction efficiency of 75.8% and uniformity of 5.8% are measured. We see that the simulation and experiment results are fairly good agreement with each other. In this paper, Genetic Algorithm and Neural Network have been successfully combined in designing CGHs. This method gives a significant reduction in computation time compared to the GA method while still allowing holograms of high diffraction efficiency and uniformity to be achieved. This work was supported by No.mOl-2001-000-00324-0 (2002)) from the Korea Science & Engineering Foundation.

• Journal of Internet Computing and Services
• /
• v.23 no.5
• /
• pp.145-154
• /
• 2022

Automatic Target Recognition (ATR) technology is emerging as a core technology of Future Combat Systems (FCS). Conventional ATR is performed based on IMINT (image information) collected from the SAR sensor, and various image-based deep learning models are used. However, with the development of IT and sensing technology, even though data/information related to ATR is expanding to HUMINT (human information) and SIGINT (signal information), ATR still contains image oriented IMINT data only is being used. In complex and diversified battlefield situations, it is difficult to guarantee high-level ATR accuracy and generalization performance with image data alone. Therefore, we propose a knowledge graph-based ATR method that can utilize image and text data simultaneously in this paper. The main idea of the knowledge graph and deep model-based ATR method is to convert the ATR image and text into graphs according to the characteristics of each data, align it to the knowledge graph, and connect the heterogeneous ATR data through the knowledge graph. In order to convert the ATR image into a graph, an object-tag graph consisting of object tags as nodes is generated from the image by using the pre-trained image object recognition model and the vocabulary of the knowledge graph. On the other hand, the ATR text uses the pre-trained language model, TF-IDF, co-occurrence word graph, and the vocabulary of knowledge graph to generate a word graph composed of nodes with key vocabulary for the ATR. The generated two types of graphs are connected to the knowledge graph using the entity alignment model for improvement of the ATR performance from images and texts. To prove the superiority of the proposed method, 227 documents from web documents and 61,714 RDF triples from dbpedia were collected, and comparison experiments were performed on precision, recall, and f1-score in a perspective of the entity alignment..

• Korean Journal of Remote Sensing
• /
• v.39 no.5_4
• /
• pp.1173-1183
• /
• 2023

Over the past 20 years, Korea's overall import and export cargo volume has increased at an average annual rate of approximately 5.3%. About 99% of the cargo is still being transported by sea. Due to recent increases in maritime cargo volume, congestion in maritime logistics has become challenging due to factors such as the COVID-19 pandemic and conflicts. Continuous monitoring of ports has become crucial. Various ground observation systems and Automatic Identification System (AIS) data have been utilized for monitoring ports and conducting numerous preliminary studies for the efficient operation of container terminals and cargo volume prediction. However, small and developing countries' ports face difficulties in monitoring due to environmental issues and aging infrastructure compared to large ports. Recently, with the increasing utility of artificial satellites, preliminary studies have been conducted using satellite imagery for continuous maritime cargo data collection and establishing ocean monitoring systems in vast and hard-to-reach areas. This study aims to visually detect ships docked at berths in the Busan New Port using high-resolution satellite imagery and quantitatively evaluate berth utilization rates. By utilizing high-resolution satellite imagery from Compact Advanced Satellite 500-1 (CAS500-1), Korea Multi-Purpose satellite-3 (KOMPSAT-3), PlanetScope, and Sentinel-2A, ships docked within the port berths were visually detected. The berth utilization rate was calculated using the total number of ships that could be docked at the berths. The results showed variations in berth utilization rates on June 2, 2022, with values of 0.67, 0.7, and 0.59, indicating fluctuations based on the time of satellite image capture. On June 3, 2022, the value remained at 0.7, signifying a consistent berth utilization rate despite changes in ship types. A higher berth utilization rate indicates active operations at the berth. This information can assist in basic planning for new ship operation schedules, as congested berths can lead to longer waiting times for ships in anchorages, potentially resulting in increased freight rates. The duration of operations at berths can vary from several hours to several days. The results of calculating changes in ships at berths based on differences in satellite image capture times, even with a time difference of 4 minutes and 49 seconds, demonstrated variations in ship presence. With short observation intervals and the utilization of high-resolution satellite imagery, continuous monitoring within ports can be achieved. Additionally, utilizing satellite imagery to monitor changes in ships at berths in minute increments could prove useful for small and developing country ports where harbor management is not well-established, offering valuable insights and solutions.

• Korean Journal of Environmental Biology
• /
• v.39 no.1
• /
• pp.119-135
• /
• 2021

Korea has stepped up efforts to investigate and catalog its flora and fauna to conserve the biodiversity of the Korean Peninsula and secure biological resources since the ratification of the Convention on Biological Diversity (CBD) in 1992 and the Nagoya Protocol on Access to Genetic Resources and the Fair and Equitable Sharing of Benefits (ABS) in 2010. Thus, after its establishment in 2007, the National Institute of Biological Resources (NIBR) of the Ministry of Environment of Korea initiated a project called the Korean Indigenous Species Investigation Project to investigate indigenous species on the Korean Peninsula. For 15 years since its beginning in 2006, this project has been carried out in five phases, Phase 1 from 2006-2008, Phase 2 from 2009-2011, Phase 3 from 2012-2014, Phase 4 from 2015-2017, and Phase 5 from 2018-2020. Before this project, in 2006, the number of indigenous species surveyed was 29,916. The figure was cumulatively aggregated at the end of each phase as 33,253 species for Phase 1 (2008), 38,011 species for Phase 2 (2011), 42,756 species for Phase 3 (2014), 49,027 species for Phase 4 (2017), and 54,428 species for Phase 5(2020). The number of indigenous species surveyed grew rapidly, showing an approximately 1.8-fold increase as the project progressed. These statistics showed an annual average of 2,320 newly recorded species during the project period. Among the recorded species, a total of 5,242 new species were reported in scientific publications, a great scientific achievement. During this project period, newly recorded species on the Korean Peninsula were identified using the recent taxonomic classifications as follows: 4,440 insect species (including 988 new species), 4,333 invertebrate species except for insects (including 1,492 new species), 98 vertebrate species (fish) (including nine new species), 309 plant species (including 176 vascular plant species, 133 bryophyte species, and 39 new species), 1,916 algae species (including 178 new species), 1,716 fungi and lichen species(including 309 new species), and 4,812 prokaryotic species (including 2,226 new species). The number of collected biological specimens in each phase was aggregated as follows: 247,226 for Phase 1 (2008), 207,827 for Phase 2 (2011), 287,133 for Phase 3 (2014), 244,920 for Phase 4(2017), and 144,333 for Phase 5(2020). A total of 1,131,439 specimens were obtained with an annual average of 75,429. More specifically, 281,054 insect specimens, 194,667 invertebrate specimens (except for insects), 40,100 fish specimens, 378,251 plant specimens, 140,490 algae specimens, 61,695 fungi specimens, and 35,182 prokaryotic specimens were collected. The cumulative number of researchers, which were nearly all professional taxonomists and graduate students majoring in taxonomy across the country, involved in this project was around 5,000, with an annual average of 395. The number of researchers/assistant researchers or mainly graduate students participating in Phase 1 was 597/268; 522/191 in Phase 2; 939/292 in Phase 3; 575/852 in Phase 4; and 601/1,097 in Phase 5. During this project period, 3,488 papers were published in major scientific journals. Of these, 2,320 papers were published in domestic journals and 1,168 papers were published in Science Citation Index(SCI) journals. During the project period, a total of 83.3 billion won (annual average of 5.5 billion won) or approximately US $75 million (annual average of US $5 million) was invested in investigating indigenous species and collecting specimens. This project was a large-scale research study led by the Korean government. It is considered to be a successful example of Korea's compressed development as it attracted almost all of the taxonomists in Korea and made remarkable achievements with a massive budget in a short time. The results from this project led to the National List of Species of Korea, where all species were organized by taxonomic classification. Information regarding the National List of Species of Korea is available to experts, students, and the general public (https://species.nibr.go.kr/index.do). The information, including descriptions, DNA sequences, habitats, distributions, ecological aspects, images, and multimedia, has been digitized, making contributions to scientific advancement in research fields such as phylogenetics and evolution. The species information also serves as a basis for projects aimed at species distribution and biological monitoring such as climate-sensitive biological indicator species. Moreover, the species information helps bio-industries search for useful biological resources. The most meaningful achievement of this project can be in providing support for nurturing young taxonomists like graduate students. This project has continued for the past 15 years and is still ongoing. Efforts to address issues, including species misidentification and invalid synonyms, still have to be made to enhance taxonomic research. Research needs to be conducted to investigate another 50,000 species out of the estimated 100,000 indigenous species on the Korean Peninsula.

• The Journal of Korean Society for Radiation Therapy
• /
• v.31 no.2
• /
• pp.43-49
• /
• 2019

Purpose: This study examined changes in the position of the heat and lungs depending on the patient's breathing method during left breast cancer radiotherapy and used treatment plans to compare the resulting radiation dose. Materials and methods: The participants consisted of 10 patients with left breast cancer. A CT simulator(SIMENS SOMATOM AS, Germany) was used to obtain images when using three different breathing methods: free breathing(FB), deep inspiration breath hold(DIBH with Abches, DIBH), inspiration breath hold(IBH with CPAP, CPAP). A Ray Station(5.0.2.35, Sweden) was used for treatment planning, the treatment method was volumetric modulated arc therapy (VMAT) with one partial arc of the same angle, and the prescribed dose to the planning target volume (PTV) was a total dose of 50Gy(2Gy/day). In treatment plan analysis, the 95% dose (D95) to the PTV, the conformity index(CI), and the homogeneity index (HI) were compared. The lungs, heart, and left anterior descending artery (LAD) were selected as the organs at risk(OARs). Results: The mean volume of the ipsilateral lung for FB, DIBH, and CPAP was 1245.58±301.31㎤, 1790.09±362.43 ㎤, 1775.44±476.71 ㎤. The mean D95 for the PTV was 46.67±1.89Gy, 46.85±1.72Gy, 46.97±23.4Gy, and the mean CI and HI were 0.95±0.02, 0.96±0.02, 0.95±0.02 and 0.91±0.01, 0.90±0.01, 0.92±0.02. The V20 of Whole Lung was 10.74±4.50%, 8.29±3.14%, 9.12±3.29% and The V20 of the ipsilateral lung was 20.45±8.65%, 17.18±7.04%, 18.85±7.85%, the Dmean of the heart was 7.82±1.27Gy, 6.10±1.27Gy, 5.67±1.56Gy, and the Dmax of the LAD was 20.41±7.56Gy, 14.88±3.57Gy, 14.96±2.81Gy. The distance from the thoracic wall to the LAD was measured to be 11.33±4.70mm, 22.40±6.01mm, 20.14±6.23mm. Conclusion: During left breast cancer radiotherapy, the lung volume was 46.24% larger for DIBH than for FB, and 43.11% larger for CPAP than FB. The larger lung volume increases the distance between the thoracic wall and the heart. In this way, the LAD, which is one of the nearby OARs, can be more effectively protected while still satisfying the treatment plan. The lung volume was largest for DIBH, and the distance between the LAD and thoracic wall was also the greatest. However, when performing treatment with DIBH, the intra-fraction error cannot be ignored. Moreover, communication between the patient and the radiotherapist is also an important factor in DIBH treatment. When communication is problematic, or if the patient has difficulty holding their breath, we believe that CPAP could be used as an alternative to DIBH. In order to verify the clinical efficacy of CPAP, it will be necessary to perform long-term follow-up of a greater number of patients.

• Journal of the Korean Institute of Traditional Landscape Architecture
• /
• v.40 no.2
• /
• pp.1-13
• /
• 2022

This paper examined the history and development of Gyeongbokgung Palace's back garden based on historical materials and drawings such as Joseon Ilgi(Diaries of Joseon Dynasty), Joseon Wangjo Sillok(the Annals of the Joseon Dynasty), Doseongdaejido(the Great Map of Seoul), Bukgwoldohyeong(Drawing Plan of the Northern Palace), the Bukgung Palace Restoration Plan, Restoration Planning of Gyeongbokgung Palace and the following results were derived. First, it was confirmed that the Back Garden of Gyeongbokgung Palace was famous for its great location since the Goryeo Dynasty, and that it was named Namkyeong at that time and was a place where a shrine was built, and that castles and palaces were already built during the Goryeo Dynasty under the influence of Fengshui-Docham(風水圖讖) and Zhouli·Kaogongji(周禮考工記). Although the back garden of Gyeongbokgung Palace in the early Joseon Dynasty stayed out of the limelight as a back garden for the palace, it has a place value as a living space for the head of the state from King Gojong to the present. Second, in order to clearly identify the boundaries of back garden, through literature such as map of Doseongdo (Map of the Capital), La Coree, Gyeongmudae Area, Japanese Geography Custom Compendium, Korean Photo Album, JoseonGeonchukdoJip(The Illustration Book of Joseon Construction), Urban Planning Survey of Gyeongseong, it was confirmed that the current Blue House area outside Sinmumun Gate was built outside the precincts of Gyeongbokgung Palace. It was found that the area devastated through the Japanese Invasion of Korea in 1592, was used as a space where public corporations were combined through the process of reconstruction during the King Gojong period. In Japanese colonial era, the place value as a back garden of the primary palace was damaged, as the palace buildings of the back garden was relocated or destroyed, but after liberation, it was used as the presidential residence and restored the place value of the ruler. Third, in the back garden of Gyeongbokgung Palace, spatial changes proceeded through the Japanese Invasion and Japanese colonial era. The place with the greatest geographical change was Gyeongnongjae area, where the residence of the Japanese Government-General of Korea was built, and there were frequent changes in the use of the land. On the other hand, the current Gyeongmudae area, the forests next to the small garden, and the forests of Baekak were preserved in the form of traditional forests. To clarify this, 1:1200 floor plan of inner Gyeongmudae residence and satellite images were overlapped based on Sinmumun Gate, and as a result, it was confirmed that the water path originating from Baekak still exists today and the forest area did not change. Fourth, in the areas where the traditional forest landscape was inherited, the functional changes in the topography were little, and major old-age colonies are maintained. The old trees identified in this area were indicator tree species with historical value. Representatively, Pinus densiflora for. multicaulis Uyeki, located in Nokjiwon Garden, is presumed to have been preserved as one of Pinus densiflora for. multicaulis Uyeki planted next to Yongmundang, and has a historicality that has been used as a photo zone at dinners for heads of state and important guests. Lastly, in order to continuously preserve and manage the value of Gyeongbokgung Palace in Blue House, it is urgent to clarify the space value through excavation of historical materials in Japanese colonial era and establish a hierarchy of garden archaeology by era. In addition, the basis for preserving the historical landscape from the Joseon Dynasty to the modern era from Gyeongbokgung Palace should not damage the area of the old giant trees, which has been perpetuated since the past, and a follow-up study is needed to investigate all the forests in Blue House.

• Asia Marketing Journal
• /
• v.12 no.2
• /
• pp.155-175
• /
• 2010

Most of the service industries including finance and telecommunications have become matured and saturated. The competitions have become severe while the differences among brands become smaller. Therefore maintaining good relationships with customers has been critical for the service providers. In case of credit card and debit card, the similar patterns are shown. It is important for them to maintain good relationships with customers, and therefore, they have used marketing program which provides customized services to customers and utilizes the membership programs. Not only do they build and maintain good relationships, but also highlight their brands from the emotional aspects. For example, KB Card or Hyundai Card uses well-known designers' works for their credit card design. As well, they differentiate the designs of credit cards to stress on their brand personalities. BC Card introduced the credit card with perfume that a customer would like. Even though the credit card is small and not shown to public easily, it becomes more important for those companies to touch the customers' feelings with the brand personalities and their images. This is partly because of changes in consumers' lifestyles. Y-generations becomes highly likely to express themselves in many different ways and more emotional than X-generations. For the Y-generations, therefore, even credit cards in the wallet should be personalized and well-designed. In line with it, credit cards with good design can be seen as an example of brand identity, where different design for each customer can be used to recognize the membership groups that customers want to belong. On the other hand, these credit card companies offer the special treatment benefits for those customers who are heavy users for the cards. For example, those customers who love sports will receive some special discounts when they use their credit cards for sports related products. Therefore this study attempted to explore the relationships between relational benefits, brand identification and loyalty. It has been well known that relational benefits and brand identification lead to loyalty independently from many other studies, but there has been few study to review all the three variables all together in a research model. Furthermore, as reviewed above, in the card industry, many companies attempt to associate the brand image with their products to fit their customers' lifestyles while relational benefits are still playing an important role for their business. Therefore in our research model, relational benefits, brand identification, and loyalty are all included. We focus on the mediating effect of brand identification. From the relational benefits perspective, only special treatment benefit and confidence benefit are included. Social benefit is not applicable for this credit card industry because not many cases of face-to-face interaction can be found. From the brand identification perspective, personal brand identity and social brand identity are reviewed and included in the model. Overall, the research model emphasizes that the relationships between relational benefits and loyalty will be mediated by the effect of brand identification. The effects of relational benefits which are confidence benefit and special treatment benefits on loyalty will be realized when they fit to the personal brand identity and social brand identity. In the research model, therefore, the relationships between confidence benefit and social brand identity, and between confidence benefit and personal identity are hypothesized while the effects of special treatment benefit on social brand identity and personal brand identity are hypothesized. Loyalty, then, is hypothesized to have positive relationships with personal brand identity and social brand identity. In addition, confidence benefit among the relational benefits is expected to have a direct, positive relationship with loyalty because confidence benefit has been recognized as a critical factor for good relationships and satisfaction. Data were collected from college students who have been using either credit cards or debit cards. College students were regarded good subjects because they are in Y-generation cohorts and have tendency to express themselves more. Total sample size was two hundred three at the beginning, but after deleting those data with many missing values, one hundred ninety-seven data points were remained and used for the model testing. Measurement items were brought from the previous literatures and modified for this research. To test the reliability, using SPSS 14, chronbach's α was examined and all the values were from .874 to .928 exceeding over .7. Using AMOS 7.0, confirmatory factor analysis was conducted to investigate the measurement model. The measurement model was found good fit with χ²(67)=188.388 (p= .000), GFI=.886, AGFI=.821, CFI=.941, RMSEA=.096. Using AMOS 7.0, structural equation modeling has been used to analyze the research model. Overall, the research model fit were χ²(68)=188.670 (p= .000), GFI=.886, AGFI=,824 CFI=.942, RMSEA=.095 indicating good fit. In details, all the paths hypothesized in the research model were found significant except for the path from social brand identity to loyalty. Personal brand identity leads to loyalty while both confidence benefit and special treatment benefit have a positive relationships with personal and social identities. As well, confidence benefit has a direct positive effect on loyalty. The results indicates the followings. First, personal brand identity plays an important role for credit/debit card usage. Therefore even for the products which are not shown to public easy, design and emotional aspect can be important to fit the customers' lifestyles. Second, confidence benefit and special treatment benefit have a positive effects on personal brand identity. Therefore it will be needed for marketers to associate the special treatment and trust and confidence benefits with personal image, personality and personal identity. Third, this study found again the importance of confidence and trust. However interestingly enough, social brand identity was not found to be significantly related to loyalty. It can be explained that the main sample of this study consists of college students. Those strategies to facilitate social brand identity are focused on high social status groups while college students have not been established their status yet.

• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.205-225
• /
• 2018

Convolutional Neural Network (ConvNet) is one class of the powerful Deep Neural Network that can analyze and learn hierarchies of visual features. Originally, first neural network (Neocognitron) was introduced in the 80s. At that time, the neural network was not broadly used in both industry and academic field by cause of large-scale dataset shortage and low computational power. However, after a few decades later in 2012, Krizhevsky made a breakthrough on ILSVRC-12 visual recognition competition using Convolutional Neural Network. That breakthrough revived people interest in the neural network. The success of Convolutional Neural Network is achieved with two main factors. First of them is the emergence of advanced hardware (GPUs) for sufficient parallel computation. Second is the availability of large-scale datasets such as ImageNet (ILSVRC) dataset for training. Unfortunately, many new domains are bottlenecked by these factors. For most domains, it is difficult and requires lots of effort to gather large-scale dataset to train a ConvNet. Moreover, even if we have a large-scale dataset, training ConvNet from scratch is required expensive resource and time-consuming. These two obstacles can be solved by using transfer learning. Transfer learning is a method for transferring the knowledge from a source domain to new domain. There are two major Transfer learning cases. First one is ConvNet as fixed feature extractor, and the second one is Fine-tune the ConvNet on a new dataset. In the first case, using pre-trained ConvNet (such as on ImageNet) to compute feed-forward activations of the image into the ConvNet and extract activation features from specific layers. In the second case, replacing and retraining the ConvNet classifier on the new dataset, then fine-tune the weights of the pre-trained network with the backpropagation. In this paper, we focus on using multiple ConvNet layers as a fixed feature extractor only. However, applying features with high dimensional complexity that is directly extracted from multiple ConvNet layers is still a challenging problem. We observe that features extracted from multiple ConvNet layers address the different characteristics of the image which means better representation could be obtained by finding the optimal combination of multiple ConvNet layers. Based on that observation, we propose to employ multiple ConvNet layer representations for transfer learning instead of a single ConvNet layer representation. Overall, our primary pipeline has three steps. Firstly, images from target task are given as input to ConvNet, then that image will be feed-forwarded into pre-trained AlexNet, and the activation features from three fully connected convolutional layers are extracted. Secondly, activation features of three ConvNet layers are concatenated to obtain multiple ConvNet layers representation because it will gain more information about an image. When three fully connected layer features concatenated, the occurring image representation would have 9192 (4096+4096+1000) dimension features. However, features extracted from multiple ConvNet layers are redundant and noisy since they are extracted from the same ConvNet. Thus, a third step, we will use Principal Component Analysis (PCA) to select salient features before the training phase. When salient features are obtained, the classifier can classify image more accurately, and the performance of transfer learning can be improved. To evaluate proposed method, experiments are conducted in three standard datasets (Caltech-256, VOC07, and SUN397) to compare multiple ConvNet layer representations against single ConvNet layer representation by using PCA for feature selection and dimension reduction. Our experiments demonstrated the importance of feature selection for multiple ConvNet layer representation. Moreover, our proposed approach achieved 75.6% accuracy compared to 73.9% accuracy achieved by FC7 layer on the Caltech-256 dataset, 73.1% accuracy compared to 69.2% accuracy achieved by FC8 layer on the VOC07 dataset, 52.2% accuracy compared to 48.7% accuracy achieved by FC7 layer on the SUN397 dataset. We also showed that our proposed approach achieved superior performance, 2.8%, 2.1% and 3.1% accuracy improvement on Caltech-256, VOC07, and SUN397 dataset respectively compare to existing work.

• MISULJARYO - National Museum of Korea Art Journal
• /
• v.96
• /
• pp.123-154
• /
• 2019

Ancestral Burial Ground on the Inwangsan Mountain is a ten-panel folding screen with images and postscripts. Commissioned by Bak Gyeong-bin (dates unknown), this screen was painted by Jo Jung-muk (1820-after 1894) in 1868. The postscripts were written by Hong Seon-ju (dates unknown). The National Museum of Korea restored this painting, which had been housed in the museum on separate sheets, to its original folding screen format. The museum also opened the screen to the public for the first time at the special exhibition Through the Eyes of Joseon Painters: Real Scenery Landscapes of Korea held from July 23 to September 22, 2019. Ancestral Burial Ground on the Inwangsan Mountain depicts real scenery on the western slopes of Inwangsan Mountain spanning present-day Hongje-dong and Hongeun-dong in Seodaemun-gu, Seoul. In the distance, the Bukhansan Mountain ridges are illustrated. The painting also bears place names, including Inwangsan Mountain, Chumohyeon Hill, Hongjewon Inn, Samgaksan Mountain, Daenammun Gate, and Mireukdang Hall. The names and depictions of these places show similarities to those found on late Joseon maps. Jo Jung-muk is thought to have studied the geographical information marked on maps so as to illustrate a broad landscape in this painting. Field trips to the real scenery depicted in the painting have revealed that Jo exaggerated or omitted natural features and blended and arranged them into a row for the purposes of the horizontal picture plane. Jo Jung-muk was a painter proficient at drawing conventional landscapes in the style of the Southern School of Chinese painting. Details in Ancestral Burial Ground on the Inwangsan Mountain reflect the painting style of the School of Four Wangs. Jo also applied a more decorative style to some areas. The nineteenth-century court painters of the Dohwaseo(Royal Bureau of Painting), including Jo, employed such decorative painting styles by drawing houses based on painting manuals, applying dots formed like sprinkled black pepper to depict mounds of earth and illustrating flowers by dotted thick pigment. Moreover, Ancestral Burial Ground on the Inwangsan Mountain shows the individualistic style of Jeong Seon(1676~1759) in the rocks drawn with sweeping brushstrokes in dark ink, the massiveness of the mountain terrain, and the pine trees simply depicted using horizontal brushstrokes. Jo Jung-muk is presumed to have borrowed the authority and styles of Jeong Seon, who was well-known for his real scenery landscapes of Inwangsan Mountain. Nonetheless, the painting lacks an spontaneous sense of space and fails in conveying an impression of actual sites. Additionally, the excessively grand screen does not allow Jo Jung-muk to fully express his own style. In Ancestral Burial Ground on the Inwangsan Mountain, the texts of the postscripts nicely correspond to the images depicted. Their contents can be divided into six parts: (1) the occupant of the tomb and the reason for its relocation; (2) the location and geomancy of the tomb; (3) memorial services held at the tomb and mysterious responses received during the memorial services; (4) cooperation among villagers to manage the tomb; (5) the filial piety of Bak Gyeong-bin, who commissioned the painting and guarded the tomb; and (6) significance of the postscripts. The second part in particular is faithfully depicted in the painting since it can easily be visualized. According to the fifth part revealing the motive for the production of the painting, the commissioner Bak Gyeongbin was satisfied with the painting, stating that "it appears impeccable and is just as if the tomb were newly built." The composition of the natural features in a row as if explaining each one lacks painterly beauty, but it does succeed in providing information on the geomantic topography of the gravesite. A fair number of the existing depictions of gravesites are woodblock prints of family gravesites produced after the eighteenth century. Most of these are included in genealogical records and anthologies. According to sixteenth- and seventeenth-century historical records, hanging scrolls of family gravesites served as objects of worship. Bowing in front of these paintings was considered a substitute ritual when descendants could not physically be present to maintain their parents' or other ancestors' tombs. Han Hyo-won (1468-1534) and Jo Sil-gul (1591-1658) commissioned the production of family burial ground paintings and asked distinguished figures of the time to write a preface for the paintings, thus showing off their filial piety. Such examples are considered precedents for Ancestral Burial Ground on the Inwangsan Mountain. Hermitage of the Recluse Seokjeong in a private collection and Old Villa in Hwagae County at the National Museum of Korea are not paintings of family gravesites. However, they serve as references for seventeenth-century paintings depicting family gravesites in that they are hanging scrolls in the style of the paintings of literary gatherings and they illustrate geomancy. As an object of worship, Ancestral Burial Ground on the Inwangsan Mountain recalls a portrait. As indicated in the postscripts, the painting made Bak Gyeong-bin "feel like hearing his father's cough and seeing his attitudes and behaviors with my eyes." The fable of Xu Xiaosu, who gazed at the portrait of his father day and night, is reflected in this gravesite painting evoking a deceased parent. It is still unclear why Bak Gyeong-bin commissioned Ancestral Burial Ground on the Inwangsan Mountain to be produced as a real scenery landscape in the folding screen format rather than a hanging scroll or woodblock print, the conventional formats for a family gravesite paintings. In the nineteenth century, commoners came to produce numerous folding screens for use during the four rites of coming of age, marriage, burial, and ancestral rituals. However, they did not always use the screens in accordance with the nature of these rites. In the Ancestral Burial Ground on the Inwangsan Mountain, the real scenery landscape appears to have been emphasized more than the image of the gravesite in order to allow the screen to be applied during different rituals or for use to decorate space. The burial mound, which should be the essence of Ancestral Burial Ground on the Inwangsan Mountain, might have been obscured in order to hide its violation of the prohibition on the construction of tombs on the four mountains around the capital. At the western foot of Inwangsan Mountain, which was illustrated in this painting, the construction of tombs was forbidden. In 1832, a tomb discovered illegally built on the forbidden area was immediately dug up and the related people were severely punished. This indicates that the prohibition was effective until the mid-nineteenth century. The postscripts on the Ancestral Burial Ground on the Inwangsan Mountain document in detail Bak Gyeong-bin's efforts to obtain the land as a burial site. The help and connivance of villagers were necessary to use the burial site, probably because constructing tombs within the prohibited area was a burden on the family and villagers. Seokpajeong Pavilion by Yi Han-cheol (1808~1880), currently housed at the Los Angeles County Museum of Art, is another real scenery landscape in the format of a folding screen that is contemporaneous and comparable with Ancestral Burial Ground on the Inwangsan Mountain. In 1861 when Seokpajeong Pavilion was created, both Yi Han-cheol and Jo Jung-muk participated in the production of a portrait of King Cheoljong. Thus, it is highly probable that Jo Jung-muk may have observed the painting process of Yi's Seokpajeong Pavilion. A few years later, when Jo Jungmuk was commissioned to produce Ancestral Burial Ground on the Inwangsan Mountain, his experience with the impressive real scenery landscape of the Seokpajeong Pavilion screen could have been reflected in his work. The difference in the painting style between these two paintings is presumed to be a result of the tastes and purposes of the commissioners. Since Ancestral Burial Ground on the Inwangsan Mountain contains the multilayered structure of a real scenery landscape and family gravesite, it seems to have been perceived in myriad different ways depending on the viewer's level of knowledge, closeness to the commissioner, or viewing time. In the postscripts to the painting, the name and nickname of the tomb occupant as well as the place of his surname are not recorded. He is simply referred to as "Mister Bak." Biographical information about the commissioner Bak Gyeong-bin is also unavailable. However, given that his family did not enter government service, he is thought to have been a person of low standing who could not become a member of the ruling elite despite financial wherewithal. Moreover, it is hard to perceive Hong Seon-ju, who wrote the postscripts, as a member of the nobility. He might have been a low-level administrative official who belonged to the Gyeongajeon, as documented in the Seungjeongwon ilgi (Daily Records of Royal Secretariat of the Joseon Dynasty). Bak Gyeong-bin is presumed to have moved the tomb of his father to a propitious site and commissioned Ancestral Burial Ground on the Inwangsan Mountain to stress his filial piety, a conservative value, out of his desire to enter the upper class. However, Ancestral Burial Ground on the Inwangsan Mountain failed to live up to its original purpose and ended up as a contradictory image due to its multiple applications and the concern over the exposure of the violation of the prohibition on the construction of tombs on the prohibited area. Forty-seven years after its production, this screen became a part of the collection at the Royal Yi Household Museum with each panel being separated. This suggests that Bak Gyeong-bin's dream of bringing fortune and raising his family's social status by selecting a propitious gravesite did not come true.