• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.32B no.12
• /
• pp.1581-1590
• /
• 1995

Recently, advanced video coding techniques using segmentation technique have been actively researched as candidates for video coding of MPEG-4 standard. The conventional segmentation techniques are unsuitable for real-time process because they have sequential structure. In this paper, we propose a new image segmentation technique using competitive learning neural network for vector quantization. The proposed segmentation procedure consist of prefiltering, primary and secondary segmentation, and a small region ellimination process. Primary segmentation segments input image in detail. Secondary segmentation merges similar region using a repetitive FSCL(Frequency sensitive competive learning) neural network. In this process, it is possible to segment an image from high resolution to low resolution by adjusting the number of repetition. Finally, small regions are merged into adjacent regions. Experimental results show that the procedure described yields reconstructed images of reasonably acceptable quality at bit rates of 0. 25 - 0.3 bit/pel.

• Journal of Electrical Engineering and Technology
• /
• v.11 no.1
• /
• pp.234-240
• /
• 2016

We propose a novel framework for markerless camera-based games. By using a visual camera, our method may yield robust human body segmentation with high performance comparable to the segmentation using depth cameras. The edges of human bodies are detected by subtracting gradient backgrounds, and human body regions are segmented by the operations based on mathematical morphology. Collisions between detected regions and virtual objects are determined by finding the colliding time using intra-frame positions of virtual objects. Experimental results show that the proposed method may produce robust segmentation of human bodies, thereby and the collision responses are more accurate than previous methods. Therefore, the proposed framework can be widely used in camera-based games requiring high performance.

• Journal of Broadcast Engineering
• /
• v.27 no.7
• /
• pp.1011-1020
• /
• 2022

Recent high-performance panoptic segmentation models are based on transformer architectures. However, transformer-based panoptic segmentation methods are basically slower than convolution-based methods, since the attention mechanism in the transformer requires quadratic complexity w.r.t. image resolution. Also, sine and cosine computation for positional embedding in the transformer also yields a bottleneck for computation time. To address these problems, we adopt three modules to speed up the inference runtime of the transformer-based panoptic segmentation. First, we perform channel-level reduction using depth-wise separable convolution for inputs of the transformer decoder. Second, we replace sine and cosine-based positional encoding with convolution operations, called conv-embedding. We also apply a separable self-attention to the transformer encoder to lower quadratic complexity to linear one for numbers of image pixels. As result, the proposed model achieves 44% faster frame per second than baseline on ADE20K panoptic validation dataset, when we use all three modules.

• Korean Journal of Radiology
• /
• v.21 no.6
• /
• pp.660-669
• /
• 2020

Objective: To evaluate the accuracy of a deep learning-based automated segmentation of the left ventricle (LV) myocardium using cardiac CT. Materials and Methods: To develop a fully automated algorithm, 100 subjects with coronary artery disease were randomly selected as a development set (50 training / 20 validation / 30 internal test). An experienced cardiac radiologist generated the manual segmentation of the development set. The trained model was evaluated using 1000 validation set generated by an experienced technician. Visual assessment was performed to compare the manual and automatic segmentations. In a quantitative analysis, sensitivity and specificity were calculated according to the number of pixels where two three-dimensional masks of the manual and deep learning segmentations overlapped. Similarity indices, such as the Dice similarity coefficient (DSC), were used to evaluate the margin of each segmented masks. Results: The sensitivity and specificity of automated segmentation for each segment (1-16 segments) were high (85.5-100.0%). The DSC was 88.3 ± 6.2%. Among randomly selected 100 cases, all manual segmentation and deep learning masks for visual analysis were classified as very accurate to mostly accurate and there were no inaccurate cases (manual vs. deep learning: very accurate, 31 vs. 53; accurate, 64 vs. 39; mostly accurate, 15 vs. 8). The number of very accurate cases for deep learning masks was greater than that for manually segmented masks. Conclusion: We present deep learning-based automatic segmentation of the LV myocardium and the results are comparable to manual segmentation data with high sensitivity, specificity, and high similarity scores.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.58 no.4
• /
• pp.806-816
• /
• 2009

Character recognition system consists of four step; text localization, text segmentation, character segmentation, and recognition. The character segmentation is very important and difficult because of noise, illumination, and so on. For high recognition rates of the system, it is necessary to take good performance of character segmentation algorithm. Many algorithms for character segmentation have been developed up to now, and many people have been recently making researches in segmentation of touching or overlapping character. Most of algorithms cannot apply to the text regions of management number marked on the slab in steel image, because the text regions are irregular such as touching character by strong illumination and by trouble of nozzle in marking machine, and loss of character. It is difficult to gain high success rate in various cases. This paper describes a new algorithm of character segmentation to recognize slab management number marked on the slab in the steel image. It is very important that pre-processing step is to convert gray image to binary image without loss of character and touching character. In this binary image, non-touching characters are simply separated by using vertical projection profile. For separating touching characters, after we use combined profile to find candidate points of boundary, decide real character boundary by using method based on recognition. In recognition step, we remove noise of character images, then recognize respective character images. In this paper, the proposed algorithm is effective for character segmentation and recognition of various text regions on the slab in steel image.

• Proceedings of the IEEK Conference
• /
• 2000.11c
• /
• pp.145-148
• /
• 2000

This paper proposes a character segmentation method for Korean letter address image. The poor quality of image binarization results in broken character strokes. To overcome this problem, two steps of processing ate introduced. The first one is to merge broken characters to generate character candidates, and the other one is to reduce the complexity of segmentation graph path. These two steps do not use recognition information to keep in high-speed.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.28 no.2C
• /
• pp.147-161
• /
• 2003

Video object segmentation is one of the core technologies for content-based real-time MPEG-4 encoding system. For real-time requirement, the segmentation algorithm should be fast and accurate but almost all existing algorithms are computationally intensive and not suitable for real-time applications. The MPEG-4 VM(Verification Model) has provided basic algorithms for MPEG-4 encoding but it has many limitations in practical software development, real-time camera input system and compression efficiency. In this paper, we implemented the preprocessing system for real-time camera input and VOP extraction for content-based video coding and also implemented motion detection to achieve the 180 : 1 compression rate for real-time and high compression MPEG-4 encoding.

• ETRI Journal
• /
• v.35 no.2
• /
• pp.321-331
• /
• 2013

High-tech markets are unpredictable owing to rapid technology innovation, diverse customer needs, high competition, and other elements. Many scholars have attempted to explain the uncertainty in high-tech markets using their own various approaches. However, sufficiently clear ways to predict diverse changes and trends in high-tech markets have yet to be presented. Thus, this paper proposes a new approach model, that is, systematic market segmentation, to give more accurate information. Using an empirical dataset from the mobile handset market in the Republic of Korea, we conduct our research model consisting of three steps. First, we categorize nine basic segments. Second, we test the stability of these segments. Finally, we profile the characteristics of the customers and products. We conclude that the approach is able to offer more diagnostic information to both practitioners and scholars. It is expected to provide rich information for an appropriate marketing mix in practice.

• Smart Structures and Systems
• /
• v.31 no.4
• /
• pp.351-363
• /
• 2023

Visual structural inspections are an inseparable part of post-earthquake damage assessments. With unmanned aerial vehicles (UAVs) establishing a new frontier in visual inspections, there are major computational challenges in processing the collected massive amounts of high-resolution visual data. We propose twin deep learning models that can provide accurate high-resolution structural components and damage segmentation masks efficiently. The traditional approach to cope with high memory computational demands is to either uniformly downsample the raw images at the price of losing fine local details or cropping smaller parts of the images leading to a loss of global contextual information. Therefore, our twin models comprising Trainable Resizing for high-resolution Segmentation Network (TRS-Net) and DmgFormer approaches the global and local semantics from different perspectives. TRS-Net is a compound, high-resolution segmentation architecture equipped with learnable downsampler and upsampler modules to minimize information loss for optimal performance and efficiency. DmgFormer utilizes a transformer backbone and a convolutional decoder head with skip connections on a grid of crops aiming for high precision learning without downsizing. An augmented inference technique is used to boost performance further and reduce the possible loss of context due to grid cropping. Comprehensive experiments have been performed on the 3D physics-based graphics models (PBGMs) synthetic environments in the QuakeCity dataset. The proposed framework is evaluated using several metrics on three segmentation tasks: component type, component damage state, and global damage (crack, rebar, spalling). The models were developed as part of the 2^nd International Competition for Structural Health Monitoring.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.28B no.9
• /
• pp.719-726
• /
• 1991

The symplified segmentation method was proposed for hardware implementation based on the human visual system. The segmentation method using fuzzy logic and just noticeable difference(JND) is composed of pre-filtering, initial segmentation and post processing. Experimental coding results show that reconstructed image using the proposed method is good on visual percerption even at a high compression ratio of 30:1.