• ETRI Journal
• /
• v.46 no.1
• /
• pp.137-153
• /
• 2024

This study introduces a new approach in Korean morphological analysis combining dictionary-based techniques with Transformer-based deep learning models. The key innovation is the use of a BERT-based reranking system, significantly enhancing the accuracy of traditional morphological analysis. The method generates multiple suboptimal paths, then employs BERT models for reranking, leveraging their advanced language comprehension. Results show remarkable performance improvements, with the first-stage reranking achieving over 20% improvement in error reduction rate compared with existing models. The second stage, using another BERT variant, further increases this improvement to over 30%. This indicates a significant leap in accuracy, validating the effectiveness of merging dictionary-based analysis with contemporary deep learning. The study suggests future exploration in refined integrations of dictionary and deep learning methods as well as using probabilistic models for enhanced morphological analysis. This hybrid approach sets a new benchmark in the field and offers insights for similar challenges in language processing applications.

• Proceedings of the IEEK Conference
• /
• 1998.10a
• /
• pp.871-874
• /
• 1998

This paper presents a new morphological spatio-temporal segmentation algorithm. The algorithm incorporates intensity and motion information simultaneously, and uses morphological tools such as morphological filters and watershed algorithm. The procedure toward complete segmetnation consists of three steps: joint marker extraction, boundary decision, and motion-based region fusion. By incorporating spatial and temporal information simultaneously, we can obtain visually meaningful segmentation results. Simulation results demonstrates the efficiency of the proposed method.

• Proceedings of the Korea Society for Simulation Conference
• /
• 2001.10a
• /
• pp.95-98
• /
• 2001

Edge detection is an important problem in computer vision and image understanding. Because the threshold is difficult to properly determine, edge images gained by the usually gradient-based segmentation methods are often tend to have many disjoint or overlapping boundaries, which makes the edge images spinous. In this paper, a practical multilevel morphological filter is presented for smoothing spinous edge images. The experimental results show that the method is effective in dealing with the images of a target in the sky.

• Journal of the Korean Institute of Navigation
• /
• v.20 no.4
• /
• pp.99-109
• /
• 1996

Though median filter is used for removing noise and smoothing image. But, the result of it has distortion around edge. And then, this paper proposes new noise removing algorithm by recursive morphological processing. Basic operation is same each other, but there is some different processing method between recursive morphology and general morphology theory. This recursive morphological filter can be viewed as the weighted order static filter, and then it has a weighted SE(structuring element). Especially using this algorithm to remove the 10% gaussian noise, this paper confirmed that PSNR is improved about 0.642~1.5757 db reserving edge well better than the results of the traditional median filter.

• Journal of the Korean Society of Fashion and Beauty
• /
• v.5 no.1 s.12
• /
• pp.71-77
• /
• 2007

This study attempted to comparative analysis about the physical and morphological damage degrees of hair according to the bleaching methods by $H_2O_2$ concentration. As a result of the changes in the physical characteristics of the hair according to the deceleration operation methods by 3-6-9% and 9% $H_2O_2$ concentration, the intensity of tension considerably decreased but the elongation was increased. And the more the damage of hair, the larger the degree of change. Regarding the method of deceleration operation, the 3-6-9% deceleration operation method was more damaging than the 9% single deceleration operation method. In the change of the characteristics of the form of the hair, the control group of the 5-level deceleration hair generally showed a smooth external appearance and regular and closely overlapped epidermis and undamaged cuticle were observed. But in the 3-6-9% deceleration operation method and 9% single deceleration operation method, damages such as the irregular form of the edge of the epidermis and the unclear boundary between the epidermis occurred.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2004.05b
• /
• pp.557-562
• /
• 2004

A surface-piercing barrier model is presented for understanding morphological development in the sheltered region and investigating the main factors causing the severe accumulation. Surface-piercing structures like vertical barriers, surface docks and floating breakwaters are recently favored from the point of view of a marine scenario since they do not in general partition the natural sea. The numerical solutions are compared with experimental data on wave profiles and morphological change rates within a rectangular harbor of a constant depth protected by surface-piercing thin breakwaters as a simplified problem. Our numerical study involves several modules: 1) wave dynamics analyzed by a plane-wave approximation, 2) suspended sediment transport combined with sediment erosion-deposition model, and 3) concurrent morphological changes. Scattering waves are solved by using a plane wave method without inclusion of evanescent modes. Evanescent modes are only considered in predicting the reflection ratio against the vertical barrier and energy losses due to vortex shedding from the lower edge of plate are taken into account. A new relationship to relate the near-bed concentration to the depth-mean concentration is presented by analyzing the vertical structure of concentration. The numerical solutions were also compared with experimental data on morphological changes within a rectangular harbor of constant water depth. Through the numerical experiments, the vortex-induced flow appears to be not ignorable in predicting the morphological changes although the immersion depth of a plate is not deep.

• Journal of the Korea Society of Computer and Information
• /
• v.17 no.10
• /
• pp.61-69
• /
• 2012

The population of highly prolific, predatory Asterias amurensis is growing sharply from year to year along the coastline of Korea, a nation surrounded by water on three sides. To make matters worse, the fact that Asterias amurensis devours living fish and shellfish has caused a heavy loss for fishermen involved in the aquaculture industry. What it all boils down to is the significance of technologies allowing one to recognize Asterias amurensis individuals using underwater images for the purpose of exterminating Asterias amurensis or identifying a change in the population of Asterias amurensis or the migration route of Asterias amurensis. An improved Asterias amurensis recognition method based on the morphological characteristics of Asterias amurensis was proposed in this paper. The proposed recognition method aimed at cases marked by the lack of extraction information on concaveness and convexity, which are the morphological characteristics of Asterias amurensis. Extracting all the characteristics of Asterias amurensis from images taken underwater is very difficult. In this respect, the proposed recognition is effective in terms of recognizing individuals in a diversity of Asterias amurensis images. As a result of the experiment, Our proposed method has achieved superior performance with 92.5% than other method.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.9 no.3
• /
• pp.684-689
• /
• 2005

In this paper, we propose car license plate recognition using morphological information and an enhanced neural network. Morphological information on horizontal and vertical edges was used to extract the license plate from a car image. We used a contour tracking algorithm combined with the method of histogram and location information to extract individual characters in the extracted plate. The enhanced neural network is proposed for recognizing them, which has the method of combining the ART-1 and the supervised teaming method. The proposed method has applied to real world car images. The experimental results show that the proposed method has better the extraction rates than the methods with information of the thresholding, the RGB and the HSI, respectively. And the proposed neural network has better recognition performance than the conventional neural networks.

• Proceedings of the KSME Conference
• /
• 2007.05b
• /
• pp.2923-2928
• /
• 2007

The study of model for velocity and turbulence within the urban canopy was carried out. To evaluate existing urban model we conducted wind tunnel experiment and large-eddy simulation (LES). Mean velocity profile and turbulence are measured within simple three different obstacle arrays. To obtain supplemental data and to verify morphological model large-eddy simulation was performed. Several methods have been used to achieve embodying the flow field in urban area. Recently, morphological method obtaining flow parameters from the statistical or physical representation of obstacle elements is a arising method. It was found that all morphological model, evaluated in this study, over predict the friction velocity, most sensitive one among the flow parameters. Velocity and turbulence in the urban canopy layer were improved by the correction using 'true' friction velocity.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.7
• /
• pp.138-146
• /
• 1996

In this paper, w epropose a new mehtod to represent the shape and to recognize the object. The shape description and the matching is implemented by using the distance transform on the morphological skeleton. The employed distance transform is the chamfer (3,4) distance transform, because the chamfer distance transform (CDT) has an approximate value to the euclidean distance. The 2-D object can be represented by means of the distribution of the distance transform on the morphological skeleton, the number of skeletons, the sum of the CDT, and the other features are employed as the mtching parameters. The matching method has the invariant features (rotation, translation, and scaling), and then the method is used effectively for recognizing the differently-posed objects and/or marks of the different shape and size.