• Journal of Information Processing Systems
• /
• v.17 no.1
• /
• pp.1-13
• /
• 2021

In this research, we study the problem of font image skeletonization using an end-to-end deep adversarial network, in contrast with the state-of-the-art methods that use mathematical algorithms. Several studies have been concerned with skeletonization, but a few have utilized deep learning. Further, no study has considered generative models based on deep neural networks for font character skeletonization, which are more delicate than natural objects. In this work, we take a step closer to producing realistic synthesized skeletons of font characters. We consider using an end-to-end deep adversarial network, SkelGAN, for font-image skeletonization, in contrast with the state-of-the-art methods that use mathematical algorithms. The proposed skeleton generator is proved superior to all well-known mathematical skeletonization methods in terms of character structure, including delicate strokes, serifs, and even special styles. Experimental results also demonstrate the dominance of our method against the state-of-the-art supervised image-to-image translation method in font character skeletonization task.

• Journal of Korea Multimedia Society
• /
• v.6 no.7
• /
• pp.1312-1320
• /
• 2003

In this paper, we consider a multilayer neural network, with a single hidden layer. Error backpropagation learning method used widely in multilayer neural networks has a possibility of local minima due to the inadequate weights and the insufficient number of hidden nodes. So we propose a fuzzy supervised learning algorithm by using self-generation that self-generates hidden nodes by the compound fuzzy single layer perceptron and modified ART1. From the input layer to hidden layer, a modified ART1 is used to produce nodes. And winner take-all method is adopted to the connection weight adaptation, so that a stored pattern for some pattern gets updated. The proposed method has applied to the student identification card images. In simulation results, the proposed method reduces a possibility of local minima and improves learning speed and paralysis than the conventional error backpropagation learning algorithm.

• Nuclear Engineering and Technology
• /
• v.54 no.1
• /
• pp.310-317
• /
• 2022

The mass attenuation coefficient is the primary physical parameter to model narrow beam gamma-ray attenuation. A new machine learning based approach is proposed to model gamma-ray shielding behavior of composites alternative to theoretical calculations. Two fuzzy logic algorithms and a neural network algorithm were trained and tested with different mixture ratios of vanadium slag/epoxy resin/antimony in the 0.05 MeV-2 MeV energy range. Two of the algorithms showed excellent agreement with testing data after optimizing adjustable parameters, with root mean squared error (RMSE) values down to 0.0001. Those results are remarkable because mass attenuation coefficients are often presented with four significant figures. Different training data sizes were tried to determine the least number of data points required to train sufficient models. Data set size more than 1000 is seen to be required to model in above 0.05 MeV energy. Below this energy, more data points with finer energy resolution might be required. Neuro-fuzzy models were three times faster to train than neural network models, while neural network models depicted low RMSE. Fuzzy logic algorithms are overlooked in complex function approximation, yet grid partitioned fuzzy algorithms showed excellent calculation efficiency and good convergence in predicting mass attenuation coefficient.

• Proceedings of the KIEE Conference
• /
• 2002.11c
• /
• pp.206-209
• /
• 2002

This paper proposes a fractal dimension method for measurement of degree of similarity between prototype pattern and input pattern at ART1 (Adaptive Resonance Theory 1) neural network. In order to confirm the validity of proposed method, comparison of the performance has made between the conventional method and the proposed method through simulation. The results show that the proposed method has considerably improved the performance.

• The Journal of the Convergence on Culture Technology
• /
• v.9 no.2
• /
• pp.533-543
• /
• 2023

This study explores the influence of the COVID-19 pandemic on the Korean art market and contrasts the classic hedonic method of art price prediction with the Artificial Neural Network technique. The empirical analysis of this paper utilizes 14,639 observations of Korean art auction data from 2015 to 2021. There are three types of variables in this study: artist-related, artwork-related, and sales-related. Previous studies have suggested that these three types of variables influence art prices. The empirical findings in this research are in twofold. First, in terms of RMSE and R², the Artificial Neural Network outperforms the hedonic model. Both techniques discover that sales and artwork variables have a greater impact than artist-related attributes. Second, when the primary factors of art price are controlled, Korean art prices are found to fall dramatically in 2020, shortly following the onset of COVID-19, but to rebound in 2021. The main lesson in this study is that the Artificial Neural Network enhances art price prediction and reduces information asymmetry in the Korean art market even in the face of unanticipated turmoil such as the COVID-19 outbreak.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.11
• /
• pp.5555-5567
• /
• 2018

State-of-art instance segmentation networks are successful at generating 2D segmentation mask for region proposals with highest classification score, yet 3D object segmentation task is limited to geocentric embedding or detector of Sliding Shapes. To this end, we propose an amodal 3D instance segmentation network called A3IS-CNN, which extends the detector of Deep Sliding Shapes to amodal 3D instance segmentation by adding a new branch of 3D ConvNet called A3IS-branch. The A3IS-branch which takes 3D amodal ROI as input and 3D semantic instances as output is a fully convolution network(FCN) sharing convolutional layers with existing 3d RPN which takes 3D scene as input and 3D amodal proposals as output. For two branches share computation with each other, our 3D instance segmentation network adds only a small overhead of 0.25 fps to Deep Sliding Shapes, trading off accurate detection and point-to-point segmentation of instances. Experiments show that our 3D instance segmentation network achieves at least 10% to 50% improvement over the state-of-art network in running time, and outperforms the state-of-art 3D detectors by at least 16.1 AP.

• Journal of Intelligence and Information Systems
• /
• v.9 no.2
• /
• pp.1-18
• /
• 2003

The automatic recognition of transport containers using image processing is very hard because of the irregular size and position of identifiers, diverse colors of background and identifiers, and the impaired shapes of identifiers caused by container damages and the bent surface of container, etc. We proposed and evaluated the novel recognition algorithm of container identifiers that overcomes effectively the hardness and recognizes identifiers from container images captured in the various environments. The proposed algorithm, first, extracts the area including only all identifiers from container images by using CANNY masking and bi-directional histogram method. The extracted identifier area is binarized by the fuzzy binarization method newly proposed in this paper and by applying contour tracking method to the binarized area, container identifiers which are targets of recognition are extracted. We proposed and applied the ART2-based RBF network for recognition of container identifiers. The results of experiment for performance evaluation on the real container images showed that the proposed algorithm has more improved performance in the extraction and recognition of container identifiers than the previous algorithms.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.2 no.1
• /
• pp.43-48
• /
• 2002

This paper discusses the application of fuzzy-ARTMAP neural network to compensate the nonlinearity of satellite communication channel. The fuzzy-ARTMAP is the class of ART(adaptive resonance theory) architectures designed fur supervised loaming. It has capabilities not fecund in other neural network approaches, that includes a small number of parameters, no requirements fur the choice of initial weights, automatic increase of hidden units, and capability of adding new data without retraining previously trained data. By a match tracking process with vigilance parameter, fuzzy-ARTMAP neural network achieves a minimax teaming rule that minimizes predictive error and maximizes generalization. Thus, the system automatically leans a minimal number of recognition categories, or hidden units, to meet accuracy criteria. As a input-converting process for implementing fuzzy-ARTMAP equalizer, the sigmoid function is chosen to convert actual channel output to the proper input values of fuzzy-ARTMAP. Simulation studies are performed over satellite nonlinear channels. QPSK signals with Gaussian noise are generated at random from Volterra model. The performance of proposed fuzzy-ARTMAP equalizer is compared with MLP equalizer.

• Journal of information and communication convergence engineering
• /
• v.9 no.6
• /
• pp.637-640
• /
• 2011

The hybrid neural networks have characteristics such as fast learning times, generality, and simplicity, and are mainly used to classify learning data and to model non-linear systems. The middle layer of a hybrid neural network clusters the learning vectors by grouping homogenous vectors in the same cluster. In the clustering procedure, the homogeneity between learning vectors is represented as the distance between the vectors. Therefore, if the distances between a learning vector and all vectors in a cluster are smaller than a given constant radius, the learning vector is added to the cluster. However, the usage of a constant radius in clustering is the primary source of errors and therefore decreases the recognition success rate. To improve the recognition success rate, we proposed the enhanced hybrid network that organizes the middle layer effectively by using the enhanced ART1 network adjusting the vigilance parameter dynamically according to the similarity between patterns. The results of experiments on a large number of calling card images showed that the proposed algorithm greatly improves the character extraction and recognition compared with conventional recognition algorithms.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• v.9 no.2
• /
• pp.360-365
• /
• 2005

본 논문에서는 콘크리트 표면 균열 영상에서 균열을 효율적으로 추출하기 위한 화상처리 기법과 ART1 기반 RBF 네트워크를 제안하여 균열의 방향성을 인식한다. 본 논문에서 사용된 화상처리 기법으로는 균열 영상의 빛을 보정하기 위한 모폴로지 기법인 채움(Closing)연산을 적용하고 Sobel 마스크를 적용하여 균열 영상의 에지를 추출한 후 반복 이진화를 적용하여 균열 영상을 이진화한다. 이진화 된 영상에 두 차례에 걸쳐 잡음제거를 수행하여 콘크리트 표면 균열 영상으로부터 균열을 추출한다. 본 논문에서는 추출된 균열을 ART1 기반 RBF 네트워크에 적용하여 균열의 방향성(횡방향, 종방향, $-45^{\circ}$방향, $45^{\circ}$방향)을 자동으로 인식할 수 있는 방법을 제안한다. 제안된 ART1 기반 RBF 네트워크는 입력층과 중간층으로의 학습은 ART1을 적용하고 중간층과 출력층 간의 학습은 Delta 학습 방법을 적용한다. 실제 콘크리트 균열 영상을 적용하여 실험한 결과, 콘크리트 표면 균열 영상에서 효율적으로 균열을 추출할 수 있었고 제안된 ART1 기반 RBF 네트워크가 추출된 균열의 방향성 인식에 효율적인 것을 확인하였다.