• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.1
• /
• pp.83-88
• /
• 2024

In this research, we propose a Semi-Supervised learning based railroad surface defect detection method. The Resnet50 model, pretrained on ImageNet, was employed for the training. Data without labels are randomly selected, and then labeled to train the ResNet50 model. The trained model is used to predict the results of the remaining unlabeled training data. The predicted values exceeding a certain threshold are selected, sorted in descending order, and added to the training data. Pseudo-labeling is performed based on the class with the highest probability during this process. An experiment was conducted to assess the overall class classification performance based on the initial number of labeled data. The results showed an accuracy of 98% at best with less than 10% labeled training data compared to the overall training data.

• 제어로봇시스템학회:학술대회논문집
• /
• 2003.10a
• /
• pp.681-686
• /
• 2003

In this paper, we propose an algorithm to extract valid planar regions from stereo images for visual navigation of mobile robots. The algorithm is based on the difference image between the stereo images obtained by applying Homography matrix between stereo cameras. Illegal planar regions are filtered out by the use of labeling of the difference images and filtering of invalid blobs using the size of each blob. Also, illegal large planar regions such as walls are removed by adopting a weighted low-pass filtering of the difference image using the past difference images. The algorithms are experimented successfully by the use of stereo camera system built in a mobile robot and a PC-based real-time vision system.

• Journal of the Semiconductor & Display Technology
• /
• v.7 no.3
• /
• pp.59-63
• /
• 2008

Geo-coding refers to the process of associating data with location information, and the system deals with geographic identifiers expressed as latitude and longitude or street addresses. Although many services have been launched, there still remains a problem for users to create geo-coded photo with manually labeling GPS(Global Positioning System) coordinate or synchronizing with separate devices. In this paper, we design and implement a geo-coding system which utilizes the time and location information embedded in digital photographs in order to automatically categorize a personal photo collection. An included GPS receiver labels a photograph with its corresponding GPS coordinates, and the position of the camera is automatically recorded into the photo image header at the moment of capture. The place and time where the photo was taken allows us to provide context metadata on the management and retrieval of information.

• IEIE Transactions on Smart Processing and Computing
• /
• v.3 no.4
• /
• pp.191-194
• /
• 2014

This paper proposes a fast and efficient method to extract the skin and bone automatically in CT images. First, the images were smoothed by applying an anisotropic diffusion filter to remove noise. The whole body was then detected by thresholding, which was set automatically. In addition, the contour of the skin was segmented using morphological operators and connected component labeling (CCL). Finally, the bone was extracted by iterative thresholding.

• Progress in Medical Physics
• /
• v.30 no.2
• /
• pp.39-48
• /
• 2019

Deep learning has been applied to various medical data. In particular, current deep learning models exhibit remarkable performance at specific tasks, sometimes offering higher accuracy than that of experts for discriminating specific diseases from medical images. The current status of deep learning applications to molecular imaging can be divided into a few subtypes in terms of their purposes: differential diagnostic classification, enhancement of image acquisition, and image-based quantification. As functional and pathophysiologic information is key to molecular imaging, this review will emphasize the need for accurate biomarker acquisition by deep learning in molecular imaging. Furthermore, this review addresses practical issues that include clinical validation, data distribution, labeling issues, and harmonization to achieve clinically feasible deep learning models. Eventually, deep learning will enhance the role of theranostics, which aims at precision targeting of pathophysiology by maximizing molecular imaging functional information.

• Proceedings of the KOSOMBE Conference
• /
• v.1998 no.11
• /
• pp.189-190
• /
• 1998

In this paper, we show the realtime 3D reconstruction algorithm with the sliced CT images. The preprocessing is thresholding, labeling, contouring, and extracting dominant point. we reconstruct 3D image with dominant points using dynamic matching technique. The software implemented in Visualc++ 5.0 as a window-based application program.

• Applied Microscopy
• /
• v.44 no.3
• /
• pp.83-87
• /
• 2014

Electron tomography (ET) is a useful tool to investigate three-dimensional details based on virtual slices of relative thick specimen, and it requires complicated procedures consisted of image acquisition steps and image processing steps with computer program. Although the complicated step, this technique allows us to overcome some limitations of conventional transmission electron microscopy: (1) overlapping of information in the ultrathin section covering from 30 nm to 90 nm when we observe very small structures, (2) fragmentation of the information when we study larger structures over 100 nm. There are remarkable biological findings with ET, especially in the field of neuroscience, although it is not popular yet. Understanding of behavior of synaptic vesicle, active zone, pooling and fusion in the presynaptic terminal have been enhanced thanks to ET. Some sophisticated models of postsynaptic density with ET and immune labeling are introduced recently. In this review, we introduce principles, practical steps of ET and some recent researches in synapse biology.

• Proceedings of the KIEE Conference
• /
• 2002.07d
• /
• pp.2515-2517
• /
• 2002

The process of character recognition goes through 5 steps; image acquisition, character region extraction, preprocessing, character region segmentation, character recognition. Therefore the final recognition rate of character recognition is directly affected by the performance of each step. This paper is a leading research for object recognition using image processing algorithm which is one of the field of study in computer vision. And this paper will suggest an algorithm to extract the portion of number chain, which is part of the research embodying a system to perceive the data of manufacture and the name of the producer on the wrapping of groceries. In addition, this can extract the number chain comparatively accurate without using many complex algorithm by diving and extracting the moving number region at the same time.

• Journal of information and communication convergence engineering
• /
• v.8 no.4
• /
• pp.466-471
• /
• 2010

We have proposed a method to detect and track moving ships using position from Radar and image processor. Real-time segmentation of moving regions in image sequences is a fundamental step in the radar-camera integrated system. Algorithms for segmentation of objects are implemented by composing of background subtraction, morphologic operation, connected components labeling, region growing, and minimum enclosing rectangle. Once the moving objects are detected, tracking is only performed upon pixels labeled as foreground with reduced additional computational burdens.

• Journal of the Semiconductor & Display Technology
• /
• v.10 no.3
• /
• pp.53-59
• /
• 2011

This paper proposes a new method of traffic signal detection and recognition in an RGB color model. The proposed method firstly processes RGB-filtering in order to detect traffic signal candidates. Secondly, it performs adaptive threshold processing and then analyzes connected components of the binary image. The connected component of a traffic signal has to be satisfied with both a bounding box rate and an area rate that are defined in this paper. The traffic signal recognition system is implemented using C language in an embedded Linux system for a high-speed real-time image processing. Experiment results show that the proposed algorithms are quite successful.