• IEMEK Journal of Embedded Systems and Applications
• /
• v.15 no.2
• /
• pp.51-60
• /
• 2020

Optical character recognition (OCR) has been studied for decades because it is very useful in a variety of places. Nowadays, OCR's performance has improved significantly due to outstanding deep learning technology. Thus, there is an increasing demand for commercial-grade but affordable OCR systems. We have developed a low-cost, high-performance OCR system for the industry with the cheapest embedded developer kit that supports GPU acceleration. To achieve high accuracy for industrial use on limited computing resources, we chose a state-of-the-art text recognition algorithm that uses an end-to-end deep learning network as a baseline model. The model was then improved by replacing the feature extraction network with the best one suited to our conditions. Among the various candidate networks, EfficientNet-B3 has shown the best performance: excellent recognition accuracy with relatively low memory consumption. Besides, we have optimized the model written in TensorFlow's Python API using TensorFlow-TensorRT integration and TensorFlow's C++ API, respectively.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.10
• /
• pp.4098-4116
• /
• 2020

Mobile devices such as smartphones are very attractive targets for augmented reality (AR) services, but their limited resources make it difficult to increase the number of objects to be recognized. When the recognition process is scaled to a large number of objects, it typically requires significant computation time and memory. Therefore, most large-scale mobile AR systems rely on a server to outsource recognition process to a high-performance PC, but this limits the scenarios available in the AR services. As a part of realizing large-scale standalone mobile AR, this paper presents a solution to the problem of accuracy, memory, and speed for large-scale object recognition. To this end, we design our own basic feature and realize spatial locality, selective feature extraction, rough pose estimation, and selective feature matching. Experiments are performed to verify the appropriateness of the proposed method for realizing large-scale standalone mobile AR in terms of efficiency and accuracy.

• Smart Structures and Systems
• /
• v.29 no.1
• /
• pp.153-166
• /
• 2022

Identifying fine cracks in steel bridge facilities is a challenging task of structural health monitoring (SHM). This study proposed an end-to-end crack image segmentation framework based on a one-step Convolutional Neural Network (CNN) for pixel-level object recognition with high accuracy. To particularly address the challenges arising from small object detection in complex background, efforts were made in loss function selection aiming at sample imbalance and module modification in order to improve the generalization ability on complicated images. Specifically, loss functions were compared among alternatives including the Binary Cross Entropy (BCE), Focal, Tversky and Dice loss, with the last three specialized for biased sample distribution. Structural modifications with dilated convolution, Spatial Pyramid Pooling (SPP) and Feature Pyramid Network (FPN) were also performed to form a new backbone termed CrackDet. Models of various loss functions and feature extraction modules were trained on crack images and tested on full-scale images collected on steel box girders. The CNN model incorporated the classic U-Net as its backbone, and Dice loss as its loss function achieved the highest mean Intersection-over-Union (mIoU) of 0.7571 on full-scale pictures. In contrast, the best performance on cropped crack images was achieved by integrating CrackDet with Dice loss at a mIoU of 0.7670.

• Journal of the Korean Society of Costume
• /
• v.57 no.8
• /
• pp.89-98
• /
• 2007

Women are playing an increasing role in the society amid the increase in the age at first marriage, reduced family size, and the weakening solidarity among family members. Gold Miss is a newly coined word which reflects the change in the value of women in the wake of the individualism and pluralism amid the structural change. Gold Miss means a new X generation that is sensitive to the latest fashion and trend with high purchasing power and self-attainment goal. They do not spare any effort to invest in themselves, lead the new culture and set the cultural trend that goes beyond the simple consumption, and come into the spotlight both socially and economically. The outcome of the analysis on the Gold Miss fashion which was revealed in the mass media indicated that the fashion was the instrument to express their own images and personalities. Though they may be some difference depending on the occupation, personality, values, and others, they pursue sophisticated, intellectual, and emotional office-look that takes the trend and personality into account. In addition, they prefer business casual attire, and pursue the total fashion with perfection which uses the gorgeous bright and vivid color, daring color, accent color arrangement and accessories. The Gold Miss fashion implies the self-identity, high-end feature, and embody the symbolism of information, which the analysis on the feature and fashion of Golden Miss indicated.

• The Transactions of the Korea Information Processing Society
• /
• v.6 no.10
• /
• pp.2816-2822
• /
• 1999

In this paper, a fast thinning method for fingerprint image by separating end and bifurcation region is proposed. To detect feature points in automatic fingerprint identification system, thinning of fingerprint is essential. The end and bifurcation regions in ridge line are separated by means of run-length coding, and parallel thinning method is applied to the separated regions. The rest parts except the end and bifurcation regions are processed by connecting center points of each run. The performance of the proposed method has been evaluated by CPU processing time and thinness measurement. By the experimental results, the proposed method is fast and has high thinness value.

• ETRI Journal
• /
• v.42 no.1
• /
• pp.101-107
• /
• 2020

For Korean language processing, morphological analysis is a critical component that requires extensive work. This morphological analysis can be conducted in an end-to-end manner without requiring a complicated feature design using a sequence-to-sequence model. However, the sequence-to-sequence model has a time complexity of O(n²) for an input length n when using the attention mechanism technique for high performance. In this study, we propose a linear-time Korean morphological analysis model using a local monotonic attention mechanism relying on monotonic alignment, which is a characteristic of Korean morphological analysis. The proposed model indicates an extreme improvement in a single threaded environment and a high morphometric F1-measure even for a hard attention model with the elimination of the attention mechanism formula.

• Journal of Biomedical Engineering Research
• /
• v.44 no.3
• /
• pp.176-181
• /
• 2023

During cancer treatment, the patient's response to drugs appears differently at the cellular level. In this paper, an image-based cell phenotypic feature quantification and key feature selection method are presented to predict the response of patient-derived cancer cells to a specific drug. In order to analyze the viability characteristics of cancer cells, high-definition microscope images in which cell nuclei are fluorescently stained are used, and individual-level cell analysis is performed. To this end, first, image stitching is performed for analysis of the same environment in units of the well plates, and uneven brightness due to the effects of illumination is adjusted based on the histogram. In order to automatically segment only the cell nucleus region, which is the region of interest, from the improved image, a superpixel-based segmentation technique is applied using the fluorescence expression level and morphological information. After extracting 242 types of features from the image through the segmented cell region information, only the features related to cell viability are selected through the ReliefF algorithm. The proposed method can be applied to cell image-based phenotypic screening to determine a patient's response to a drug.

• Journal of the Korea Society of Computer and Information
• /
• v.10 no.6 s.38
• /
• pp.1-8
• /
• 2005

Facial analysis is used in many applications like face recognition systems, human-computer interface through head movements or facial expressions, model based coding, or virtual reality. In all these applications a very precise extraction of facial feature points are necessary. In this paper we presents a method for automatic extraction of the facial features Points such as mouth corners, eye corners, eyebrow corners. First, face region is detected by AdaBoost-based object detection algorithm. Then a combination of three kinds of feature energy for facial features are computed; valley energy, intensity energy and edge energy. After feature area are detected by searching horizontal rectangles which has high feature energy. Finally, a corner detection algorithm is applied on the end region of each feature area. Because we integrate three feature energy and the suggested estimation method for valley energy and intensity energy are adaptive to the illumination change, the proposed feature extraction method is robust under various conditions.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.29 no.2
• /
• pp.141-147
• /
• 2011

This study analyzed the issues and its usefulness of drawing small-scale digital map by using the large-scale digital map which was producted with high-resolution digital aerial photograph which are commonly photographed in recent years. To this end, correlation analysis of the feature categories on the digital map was conducted, and this map was processed by inputting data, organizing, deleting, editing, and supervising feature categories according to the generalization process. As a result, 18 unnecessary feature codes were deleted, and the accuracy of 1/5,000 for the digital map was met. Although the size of the data and the number of feature categories increased, this was proven to be shown due to the excellent description of the digital aerial photograph. Accordingly, it was shown that drawing a small-scale digital map with the large-scale digital map by digital aerial photograph provided excellent description and high-quality information for digital map.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.11
• /
• pp.5322-5342
• /
• 2016

In vehicle-to-vehicle (V2V) networks, where selfishness degrades node activity, countermeasures for collaboration enforcement must be provided to enable application of a sage and efficient network environment. Because vehicular networks feature both high mobility and various topologies, selfish behavior judgment and establishment of a stable routing protocol become intensely challenging. In this paper, a two-phase-based generous cooperative routing protocol (called GEC) is presented for V2V networks to provide resistance to selfishness. To detect selfish behaving vehicles, a packet forwarding watchdog and an average connection rate based on the multipath weight method are used, where evidence is gathered from different watchdogs. Then, multihop relay decisions are made using a generous cooperative algorithm based on game theory. Finally, through buffering of the multiple end-to-end paths and judicious choice of optimal cooperative routes, route maintenance phase is capable of dealing with congestion and rapidly exchanging traffic. Specifically, it is proved that the GEC is theoretically subgame perfect. Simulation results show that for V2V networks with inherently selfish nodes, the proposed method isolates uncooperative vehicles and is capable of accommodating both the mobility and congestion circumstances by facilitating information dissemination and reducing end-to-end delay.