• ETRI Journal
• /
• v.45 no.4
• /
• pp.666-677
• /
• 2023

In the last few years, convolutional neural networks (CNNs) have demonstrated good performance while solving various computer vision problems. However, since CNNs exhibit high computational complexity, signal processing is performed on the server side. To reduce the computational complexity of CNNs for edge computing, a lightweight algorithm, such as a MobileNet, is proposed. Although MobileNet is lighter than other CNN models, it commonly achieves lower classification accuracy. Hence, to find a balance between complexity and accuracy, additional hyperparameters for adjusting the size of the model have recently been proposed. However, significantly increasing the number of parameters makes models dense and unsuitable for devices with limited computational resources. In this study, we propose a novel MobileNet architecture, in which the number of parameters is adaptively increased according to the importance of feature maps. We show that our proposed network achieves better classification accuracy with fewer parameters than the conventional MobileNet.

• Journal of Broadcast Engineering
• /
• v.14 no.3
• /
• pp.322-331
• /
• 2009

This paper proposes an efficient data representation of stereo images using edge-based mesh optimization. Mash-based two dimensional warping for stereo images mainly depends on the performance of a node selection and a disparity estimation of selected nodes. Therefore, the proposed method first of all constructs the feature map which consists of both strong edges and boundary lines of objects for node selection and then generates a grid-based mesh structure using initial nodes. The displacement of each nodal position is iteratively estimated by minimizing the predicted errors between target image and predicted image after two dimensional warping for local area. Generally, iterative two dimensional warping for optimized nodal position required a high time complexity. To overcome this problem, we assume that input stereo images are only horizontal disparity and that optimal nodal position is located on the edge include object boundary lines. Therefore, proposed iterative warping method performs searching process to find optimal nodal position only on edge lines along the horizontal lines. In the experiments, we compare our proposed method with the other mesh-based methods with respect to the quality by using Peak Signal to Noise Ratio (PSNR) according to the number of nodes. Furthermore, computational complexity for an optimal mesh generation is also estimated. Therefore, we have the results that our proposed method provides an efficient stereo image representation not only fast optimal mesh generation but also decreasing of quality deterioration in spite of a small number of nodes through our experiments.

• Korean Journal of Remote Sensing
• /
• v.36 no.3
• /
• pp.449-460
• /
• 2020

Super-resolution is a technique used to reconstruct an image with low-resolution into that of high-resolution. Recently, deep-learning based super resolution has become the mainstream, and applications of these methods are widely used in the remote sensing field. In this paper, we propose a super-resolution method based on the deep back-projection network model to improve the satellite image resolution by the factor of four. In the process, we customized the loss function with the edge loss to result in a more detailed feature of the boundary of each object and to improve the stability of the model training using generative adversarial network based on Wasserstein distance loss. Also, we have applied the detail preserving image down-scaling method to enhance the naturalness of the training output. Finally, by including the modified-residual learning with a panchromatic feature in the final step of the training process. Our proposed method is able to reconstruct fine features and high frequency information. Comparing the results of our method with that of the others, we propose that the super-resolution method improves the sharpness and the clarity of WorldView-3 and KOMPSAT-2 images.

• Journal of the Korea institute for structural maintenance and inspection
• /
• v.15 no.5
• /
• pp.82-91
• /
• 2011

In this study, 24 prototype specimens were tested to find out the shear behavior and strength of large anchorage system exceeding 50mm(2") in anchor bolt diameter($d_0$) and 635mm(25") in effective embedment depth($h_{ef}$) not addressed by ACI349-06 Appendix B. Test variables are anchor bolt diameter($d_0$ = 63.5, 76.2, 88.9mm), effective embedment depth($h_{ef}$=635, 762mm), and edge distance($c_1$=381, 508, 762mm). Concrete compressive strength is constant($f_{ck}$=38MPa). Test results ($V_{test}$) were overestimated by $V_{aci06}$(shear strength by ACI 349-06) and $V_{ccd}$(shear strength by CCD method). In large anchorage system exceeding 50mm(2") of anchor bolt diameter($d_0$) and 635mm(25") of anchor bolt effective embedment depth($h_{ef}$), the bolt diameter variation and effective embedment depth($h_{ef}$) has no influence on the shear strenth, But, according to the analysis results of the feature ratio on edge distance($c_1$) and anchor bolt diameter, the feature ratio become smaller, which means anchor bolt diameter is bigger, predicted ratio of test results and predicted equation is larger. It was found that anchor bolt diameter is immediate cause of deterioration in the shear capacity of large anchorage system. To improve and extend the validity of current design recommendations further theoretical and numerical work is needed.

• Journal of Internet Computing and Services
• /
• v.7 no.2
• /
• pp.23-35
• /
• 2006

This work presents o novel method which extract facial region und features from motion picture automatically and controls the 3D facial expression in real time. To txtract facial region and facial feature points from each color frame of motion pictures a new nonparametric skin color model is proposed rather than using parametric skin color model. Conventionally used parametric skin color models, which presents facial distribution as gaussian-type, have lack of robustness for varying lighting conditions. Thus it needs additional work to extract exact facial region from face images. To resolve the limitation of current skin color model, we exploit the Hue-Tint chrominance components and represent the skin chrominance distribution as a linear function, which can reduce error for detecting facial region. Moreover, the minimal facial feature positions detected by the proposed skin model are adjusted by using edge information of the detected facial region along with the proportions of the face. To produce the realistic facial expression, we adopt Water's linear muscle model and apply the extended version of Water's muscles to variation of the facial features of the 3D face. The experiments show that the proposed approach efficiently detects facial feature points and naturally controls the facial expression of the 3D face model.

• Journal of the Korea Convergence Society
• /
• v.12 no.11
• /
• pp.45-51
• /
• 2021

With the development of deep learning, semantic segmentation methods are being studied in various fields. There is a problem that segmenation accuracy drops in fields that require accuracy such as medical image analysis. In this paper, we improved PSPNet, which is a deep learning based segmentation method to minimized the loss of features during semantic segmentation. Conventional deep learning based segmentation methods result in lower resolution and loss of object features during feature extraction and compression. Due to these losses, the edge and the internal information of the object are lost, and there is a problem that the accuracy at the time of object segmentation is lowered. To solve these problems, we improved PSPNet, which is a semantic segmentation model. The multi-scale attention proposed to the conventional PSPNet was added to prevent feature loss of objects. The feature purification process was performed by applying the attention method to the conventional PPM module. By suppressing unnecessary feature information, eadg and texture information was improved. The proposed method trained on the Cityscapes dataset and use the segmentation index MIoU for quantitative evaluation. As a result of the experiment, the segmentation accuracy was improved by about 1.5% compared to the conventional PSPNet.

• Journal of KIISE:Software and Applications
• /
• v.27 no.5
• /
• pp.551-562
• /
• 2000

Content-based image retrieval system retrieves an image from a database using visual features. Among approaches to express visual aspects in queries, 'query by sketch' is most convenient and expressive. However, every 'query by sketch' system has the query imperfectness problem. GContent-based image retrieval system retrieves an image from a database using visual features. Among approaches to express visual aspects in queries, 'query by sketch' is most convenient and expressive. However, every 'query by sketch' system has the query imperfectness problem. Generally, the query image produced by a user is different from the intended target image. To overcome this problem, many image retrieval systems use the spatial relationships of the objects, instead of pixel coordinates of the objects. In this paper, a query-converting algorithm for an image retrieval system, which uses the spatial relationship of every two objects as an image feature, is proposed. The proposed algorithm converts the query image into a graph that has the minimum number of edges, by eliminating every transitive edge. Since each edge in the graph represents the spatial relationship of two objects, the elimination of unnecessary edges makes the retrieval process more efficient. Experimental results show that the proposed algorithm leads the smaller number of comparison in searching process as compared with other algorithms that do not guarantee the minimum number of edges.

• Journal of Environmental Impact Assessment
• /
• v.17 no.5
• /
• pp.289-297
• /
• 2008

An impervious cover is one of most important factors which effect on a water body environment in a watershed. There are many researches on the impact of an impervious cover on water quality, quantity and ecosystem and most of these researches have been focused on an impervious rate or area in a watershed without considering structure features as like shape, edge, connection of impervious cover. In this study, we focused on a landscape structure which includes shape, density, contiguity, distance, aggregation of land cover type as well as area and rate. The calculation of a landscape indices made to analyse a landscape structure is conducted by applying Fragastats 3.3 program. Osan-cheon watershed where has rapidly urbanized is selected as a study field. Land information for 2002 and 2007 is from land classification maps provided by Ministry of Environment. The result shows that the increasing rate of an impervious cover is more conspicious in Kiheung dam watershed but the fragment of impervious cover areas is shown remarkably in the Osan sub-watershed. The trend of aggregation and connection of impervious covers is increasing. But it was very difficult to say that which type of landscape structure is more beneficial for a watershed management. The implication of this study is to find the need to come over the conventional ways to evaluate landscape structure of a watershed such as rates and areas of impervious cover, and define the importance of landscape feature as like connection, distance, edge density, fragment of impervious covers.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.12 no.2
• /
• pp.103-110
• /
• 2012

The population growth along with the urbanization, has caused more problems in many public areas, such as subway airport terminals, hospital, etc. Many surveillance systems have been installed in the public areas, but not all of those can be monitored in real-time, because the operators that observe the monitors are very small compared with the number of the monitors. For example, the observer can miss some crucial accidents or detect after considerable delays. Thus, intelligent surveillance system for preventing the accidents are needed, such as Intelligent Surveillance Systems. in this paper, we propose a new crowd density estimation method which aims at estimating moving crowd using images from surveillance cameras situated in outdoor locations. The moving crowd is estimated from the area where using optical flow. The edge information is also used as feature to measure the crowd density, so we improve the accuracy of estimation of crowd density. A multilayer neural network is designed to classify crowd density into 5 classes. Finally the proposed method is experimented with PETS 2009 images.

• Journal of Korea Multimedia Society
• /
• v.12 no.9
• /
• pp.1217-1232
• /
• 2009

In this paper, we propose a semi-automatic 3D building reconstruction method using uncalibrated images which includes the facade of target building. First, we extract feature points in all images and find corresponding points between each pair of images. Second, we extract lines on each image and estimate the vanishing points. Extracted lines are grouped with respect to their corresponding vanishing points. The adjacency graph is used to organize the image sequence based on the number of corresponding points between image pairs and camera calibration is performed. The initial solid model can be generated by some user interactions using grouped lines and camera pose information. From initial solid model, a detailed building model is reconstructed by a combination of predefined basic Euler operators on half-edge data structure. Automatically computed geometric information is visualized to help user's interaction during the detail modeling process. The proposed system allow the user to get a 3D building model with less user interaction by augmenting various automatically generated geometric information.