• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.3
• /
• pp.877-893
• /
• 2022

With the development of deep learning, face inpainting has been significantly enhanced in the past few years. Although image inpainting framework integrated with generative adversarial network or attention mechanism enhanced the semantic understanding among facial components, the issues of reconstruction on corrupted regions are still worthy to explore, such as blurred edge structure, excessive smoothness, unreasonable semantic understanding and visual artifacts, etc. To address these issues, we propose a Learnable Structure Knowledge of Fusion Network (LSK-FNet), which learns a prior knowledge by edge generation network for image inpainting. The architecture involves two steps: Firstly, structure information obtained by edge generation network is used as the prior knowledge for face inpainting network. Secondly, both the generated prior knowledge and the incomplete image are fed into the face inpainting network together to get the fusion information. To improve the accuracy of inpainting, both of gated convolution and region normalization are applied in our proposed model. We evaluate our LSK-FNet qualitatively and quantitatively on the CelebA-HQ dataset. The experimental results demonstrate that the edge structure and details of facial images can be improved by using LSK-FNet. Our model surpasses the compared models on L1, PSNR and SSIM metrics. When the masked region is less than 20%, L1 loss reduce by more than 4.3%.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.10a
• /
• pp.591-593
• /
• 2022

This paper proposes a partial offloading system for minimizing application service processing latency in an FEC (Fog/Edge Computing) environment, and it analyzes the offloading effect of the proposed system against local-only and edge-server-only processing based on network load. A partial offloading algorithm based on reconstruction linearization of multi-branch structures is included in the proposed system, as is an optimal collaboration algorithm between mobile devices and edge servers [1,2]. The experiment was conducted by applying layer scheduling to a logical CNN model with a DAG topology. When compared to local or edge-only executions, experimental results show that the proposed system always provides efficient task processing strategies and processing latency.

• Smart Structures and Systems
• /
• v.20 no.4
• /
• pp.441-450
• /
• 2017

The objective of this work is to present a conceptual design and the modelling of a distributed sensor system based on fiber optic devices (Fiber Bragg Grating, FBG), aimed at measuring span-wise and chord-wise variations of an adaptive (morphing) trailing edge. The network is made of two different integrated solutions for revealing deformations of the reference morphing structure. Strains are confined to typical values along the span (length) but they are expected to overcome standard ranges along the chord (width), up to almost 10%. In this case, suitable architectures may introduce proper modulations to keep the measured deformation low while preserving the information content. In the current paper, the designed monitoring system combines the use of a span-wise fiber reinforced patch with a chord-wise sliding beam. The two elements make up a closed grid, allowing the reconstruction of the complete deformed shape under the acceptable assumption that the transformation refers to regular geometry variations. Herein, the design logic and some integration issues are reported. Preliminary experimental test results are finally presented.

• The Journal of the Korea Contents Association
• /
• v.8 no.2
• /
• pp.74-82
• /
• 2008

This paper investigates chrominance reconstruction methods that reconstruct subsampled chrominance information efficiently using the correlation between luminance and chrominance components in the decompression process of compressed images, and analyzes drawbacks involved in the adaptive-weighted 2-dimensional linear interpolation among the methods, which shows higher efficiency in the view of computational complexity than other methods. To improve the drawback that the spatial frequency distribution is not considered for the decompressed image and to support the application on a low-performance system in behalf of 2-dimensional linear interpolation, this paper proposes the multi-mode reconstruction method which uses three reconstruction methods having different computational complexity from each other according to the degree of edge response of luminance component. The performance evaluation on a development platform for embedded systems showed that the proposed reconstruction method supports the similar level of image quality for decompressed images while reducing the overall computation time for chrominance reconstruction in comparison with the 2-dimensional linear interpolation.

• 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.

• Journal of Broadcast Engineering
• /
• v.4 no.1
• /
• pp.59-67
• /
• 1999

In this paper, we study two important topics in stereoscopic image communication system. One is a disparity estimation (DE) method to obtain the depth information of a scene at the transmitter and the other is an intermediate view reconstruction(IVR) method at the receiver. We propose a new DE method using an edge-preserving directional regularization technique. The proposed DE method smooths disparity vectors in smooth regions and preserves edges without over-smoothing problem. It provides better reconstructed stereoscopic images and improved coding efficiency than the existing regularization techniques. In addition. we propose a new IVR method using interpolation and extrapolation techniques. The proposed IVR method preserves edge regions as well as occlusion regions well. Thus. it gives better intermediate views than the existing IVR methods.

• Journal of the Korean Society for Precision Engineering
• /
• v.19 no.7
• /
• pp.71-79
• /
• 2002

A new rapid prototyping process, Variable Lamination Manufacturing using a 4-axis-controlled hotwire cutter and expandable polystyrene foam sheet as a laminating material of the part (VLM-S), has been developed to reduce building time and to improve the surface finish of parts. The objective of this study is to reconstruct the surface of the original 3D CAD model in order to generate mid-slice data using the advancing front technique. The generation of 3D layers by a 4 axis-controlled hot-wire cutter requires a completely different procedure to generate toolpath data unlike the conventional RP CAD systems. The cutting path data for VLM-S are created by VLM-Slicer, which is a special CAD/CAM software with automatic generation of 3D toolpath. For the conventional sheet type system like LOM, the STL file would be sliced into 2D data only. However, because of using the thick layers and a sloping edge with the firstorder approximation between the top and bottom layers, VLM-Slicer requires surface reconstruction, mid-slice, and the toolpath data generation as well as 2D slicing. Surface reconstruction demands the connection between the two neighboring cross-sectional contours using the triangular facets. VLM-S employs thick layers with finite thickness, so that surface reconstruction is necessary to obtain a sloping angle of a side surface and the point data at a half of the sheet thickness. In the process of the toolpath data generation the surface reconstruction algorithm is expected to minimize the error between the ruled surface and the original parts..

• Journal of radiological science and technology
• /
• v.47 no.3
• /
• pp.175-182
• /
• 2024

Advanced modeled iterative reconstruction (ADMIRE) represents a repetitive reconstruction method that can adjust strength and kernel, each of which are known to affect computed tomography (CT) image quality. The aim of this study was to quantitatively analyze the noise and spatial resolution of CT images according to ADMIRE control factors. Patient images were obtained by applying ADMIRE strength 2 and 3, and kernel B40 and B59. For quantitative evaluations, the noise level, spatial resolution, and overall image quality were measured using coefficient of variation (COV), edge rise distance (ERD), and natural image quality evaluation (NIQE). The superior values for the average COV, ERD, and NIQE results were obtained for the ADMIRE reconstruction conditions of ADMIRE 2 + B40, ADMIRE 3 + B59, and ADMIRE3 + B59. NIQE, which represents the overall image quality based on no-reference, was about 6.04 when using ADMIRE 3 + B59, showing the best result among the reconstructed image acquisition conditions. The results of this study indicate that the ADMIRE strength and kernel chosen for use in ADMIRE reconstruction have a significant impact on CT image quality. This highlights the importance of adjusting to the control factors in consideration of the clinical environment.

• Journal of Korea Entertainment Industry Association
• /
• v.5 no.1
• /
• pp.127-133
• /
• 2011

Characters in natural image are an important information with various context. Previous work of character region detection algorithms is not detect of character region in case of image complexity and the surrounding lighting, similar background to character, so this paper propose an method of character region detection in natural image using edge and connected component by morphological reconstructions. Firstly, we detect edge using Canny-edge detector and connected component with local min/max value by morphological reconstructed-operation in gray-scale image, and labeling each of detected connected component elements. lastly, detected candidate of text regions was merged for generation for one candidate text region, Final text region detected by checking the similarity and adjacency of neighbor of text candidate individual character. As the results of experiments, proposed algorithm improved the correctness of character regions detection using edge and connected components.

• Proceedings of the IEEK Conference
• /
• 2001.09a
• /
• pp.25-28
• /
• 2001

A theory of binary wavelets has been recently proposed by using two-band perfect reconstruction filter banks over binary field . Binary wavelet transform (BWT) of binary images can be used as an alternative to the real-valued wavelet transform of binary images in image processing applications such as compression, edge detection, and recognition. The BWT, however, requires large amount of computations since its operation is accomplished by matrix multiplication. In this paper, a fast BWT algorithm which utilizes filtering operation instead or matrix multiplication is presented . It is shown that the proposed algorithm can significantly reduce the computational complexity of the BWT. For the decomposition and reconstruction or an N ${\times}$ N image, the proposed algorithm requires only 2LN$^2$ multiplications and 2(L-1)N$^2$addtions when the filter length is L, while the BWT needs 2N$^3$multiplications and 2N(N-1)$^2$additions.