Digital Library[ Search Result ]
An Experimental Study on the Text Generation Capability for Chart Image Descriptions in Korean SLLM
http://doi.org/10.5626/JOK.2025.52.2.132
This study explores the capability of using Small Large Language Models(SLLMs) for automatically generating and interpreting information from chart images. To achieve this goal, we built an instruction dataset for SLLM training by extracting text data from chart images and adding descriptive information. We conducted instruction tuning on a Korean SLLM and evaluated its ability to generate information from chart images. The experimental results demonstrated that the SLLM, which was fine-tuned with the constructed instruction dataset, was capable of generating descriptive text comparable to OpenAI's GPT-4o-mini API. This study suggests that, in the future, Korean SLLMs may be effectively used for generating descriptive text and providing information across a broader range of visual data.
VACS: Virtual Try-on Artifact Correction System using the Fashion Object Segmentation Method
Wonjung Park, Youjin Chung, Soonchan Park, Jinah Park
http://doi.org/10.5626/JOK.2022.49.10.802
Virtual try-on (VITON) technology is receiving a lot of attention with the development of Generative Adversarial Networks (GANs) [1]. Previous approaches to VITON synthesized 2D model images and in-shop clothing images using a generative model. However, when synthesizing the top, VITON erroneously changes pixels in unintended areas, such as the background and pants. In this study, we propose the VITON Artifact Correction System (VACS), which divides and protects targeted clothes synthesized in VITON by fashion object segmentation, and replaces the pixels corresponding to the remaining areas with the original model image to increase the realism of the final composition.
Ship Detection using CNN based on Contrast Fusion Technique in Satellite Images : Accuracy Enhancement
Sunggyun Im, Youngbae Jeon, Junghwan Hwang, Jiwon Yoon
http://doi.org/10.5626/JOK.2019.46.8.823
The satellite has various missions such as ground/marine observation, communication, broadcasting, etc. Satellite photographs provide information for the maintenance of marine security and traffic control for ship detection. Since satellite photos are taken all over the earth, the memory storage is not sufficient to hold such data with each data being of a high resolution and requiring automatic ship detection using the computer. The existing literature on ship detection employed several deep learning models. However, the problem of processing speed due to the characteristics of satellite photographs leads to the necessity of using a CNN(Convolution Neural Network) model that has a comparably high processing speed. On the contrary, it is difficult to improve the accuracy and performance mostly due to factors such as marina, lighthouses and waves. Therefore, in this paper, we propose a model that improves the accuracy and performance by combining image contrast enhancement with the existing CNN. In addition, we have employed the overlap and rotation functions to increase the amount of data required for ship classification in the learning stage and implement automation detection technology considering window sliding to reduce detection speed in real satellite photographs. Also, the identified ship data has been used as learning data to improve accuracy for the model that can be used in the real industry.
Motion Area Detection Algorithm based on Irregularity of Light
http://doi.org/10.5626/JOK.2017.44.10.1094
In this paper, a motion image is detected based on the irregularity of lights. This motion image is extracted by modifying the reflected light region of the 3 way-diff algorithms. 3 way-diff algorithm extracts reflected light region using the 3-successive image. In this algorithm, The reflected light region is a region generated by light in the image production process and is finally created around all objects. The algorithm shows a process to extracting the region. This process is a simple operation, but doesn’t have a defined formula for light. This paper judges that the reflected light region is a kind of noise at the 3 way-diff algorithms and defines the formula for extracting the reflected light region. It shows that compared with the proposed algorithm and existing algorithm through experiment.
Fast and All-Purpose Area-Based Imagery Registration Using ConvNets
Together with machine-learning frameworks, area-based imagery registration techniques can be easily applied to diverse types of image pairs without predefined features and feature descriptors. However, feature detectors are often used to quickly identify candidate image patch pairs, limiting the applicability of these registration techniques. In this paper, we propose a ConvNet (Convolutional Network) “Dart“ that provides not only the matching metric between patches, but also information about their distance, which are helpful in reducing the search space of the corresponding patch pairs. In addition, we propose a ConvNet “Fad“ to identify the patches that are difficult for Dart to improve the accuracy of registration. These two networks were successfully implemented using Deep Learning with the help of a number of training instances generated from a few registered image pairs, and were successfully applied to solve a simple image registration problem, suggesting that this line of research is promising.
Topological Analysis of the Feasibility and Initial-value Assignment of Image Segmentation
This paper introduces and analyzes the theoretical basis and method of the conventional initial-value assignment problem and feasibility of image segmentation. The paper presents topological evidence and a method of appropriate initial-value assignment based on topology theory. Subsequently, the paper shows minimum conditions for feasibility of image segmentation based on separation axiom theory of topology and a validation method of effectiveness for image modeling. As a summary, this paper shows image segmentation with its mathematical validity based on topological analysis rather than statistical analysis. Finally, the paper applies the theory and methods to conventional Gaussian random field model and examines effectiveness of GRF modeling.
Fin Cutting Line Detection Technique based on RANSAC for Fish Cutting Automation System
The fishing industry requires many workers to manually carry out the jobs of sorting and cutting fishes. There are therefore many dangerous situations in their working environment and the throughput is inefficiently low. This paper introduces an automatic fin cutting system based on RANSAC that is able to increase the throughput of fish processing jobs. The system proposed in this paper first detects the edges of a fish using a high-pass filter. The boundary lines between fin and body are then detected by adjusting parameters and the threshold of the noise filters. Finally, the optimal cutting lines are detected using RANSAC. Through an experiment with a sample of 50 fishes, this paper shows that the proposed system detects the cutting lines with about 90% accuracy.
Search
Journal of KIISE
- ISSN : 2383-630X(Print)
- ISSN : 2383-6296(Electronic)
- KCI Accredited Journal
Editorial Office
- Tel. +82-2-588-9240
- Fax. +82-2-521-1352
- E-mail. chwoo@kiise.or.kr