• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.30 no.8C
• /
• pp.824-831
• /
• 2005

In the reconstruction of high resolution (HR) images from low resolution (LR) images frames, the error in the estimated motion degrades the reliability of the reconstructed HR image. Some methods were recently proposed where motion estimation and HR reconstruction is performed simultaneously. The estimated motion is still prone to error when it is based on a simple block matching. In this paper, we propose a reconstruction of a HR image by applying POCS and a regularized block matching simultaneously. In this method, a motion vector is obtained from a regularized block matching algorithm since the motion of a pixel in an image is highly correlated with the motion in neighboring regions. Experimental results show that the improved accuracy of the estimated motion vectors results in higher PSNR of the reconstructed HR images.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.11
• /
• pp.3780-3797
• /
• 2022

The estimation of radio map (RM) is a fundamental and critical task for the network planning and optimization performance of mobile communication. In this paper, a RM estimation method is proposed based on a deep dual learning structure. This method can simultaneously and accurately reconstruct the urban building map (UBM) and estimate the RM of the whole cell by only part of the measured reference signal receiving power (RSRP). Our proposed method implements UBM reconstruction task and RM estimation task by constructing a dual U-Net-based structure, which is named RadioCycle. RadioCycle jointly trains two symmetric generators of the dual structure. Further, to solve the problem of interference negative transfer in generators trained jointly for two different tasks, RadioCycle introduces a dynamic weighted averaging method to dynamically balance the learning rate of these two generators in the joint training. Eventually, the experiments demonstrate that on the UBM reconstruction task, RadioCycle achieves an F1 score of 0.950, and on the RM estimation task, RadioCycle achieves a root mean square error of 0.069. Therefore, RadioCycle can estimate both the RM and the UBM in a cell with measured RSRP for only 20% of the whole cell.

• International Journal of Control, Automation, and Systems
• /
• v.3 no.4
• /
• pp.612-619
• /
• 2005

A robust adaptive speed sensorless induction motor direct torque control (DTC) using a neural network (NN) is presented in this paper. The inherent lumped uncertainties of the induction motor DTC system such as parametric uncertainty, external load disturbance and unmodeled dynamics are approximated by the NN. An additional robust control term is introduced to compensate for the reconstruction error. A control law and adaptive laws for the weights in the NN, as well as the bounding constant of the lumped uncertainties are established so that the whole closed-loop system is stable in the sense of Lyapunov. The effect of the speed estimation error is analyzed, and the stability proof of the control system is also proved. Experimental results as well as computer simulations are presented to show the validity and efficiency of the proposed system.

• Journal of IKEEE
• /
• v.18 no.2
• /
• pp.272-281
• /
• 2014

In electrical impedance tomography (EIT), modified Newton Raphson (mNR) method is widely used inverse algorithm for static image reconstruction due to its convergence speed and estimation accuracy. The unknown conductivity distribution is estimated iteratively by minimizing a cost functional such that the residual error namely the difference in measured and calculated voltages is reduced. Although, mNR method has good estimation performance, EIT inverse problem still suffers from ill-conditioned and ill-posedness nature. To mitigate the ill-posedness, generally, regularization methods are adopted. The inverse solution is highly dependent on the choice of regularization parameter. In most cases, the regularization parameter has a constant value and is chosen based on experience or trail and error approach. In situations, when the internal distribution changes or with high measurement noise, the solution does not get converged with the use of constant regularization parameter. Therefore, in this paper, in order to improve the image reconstruction performance, we propose a new scheme to determine the regularization parameter. The regularization parameter is computed based on residual error and updated every iteration. The proposed scheme is tested with numerical simulations and laboratory phantom experiments. The results show an improved reconstruction performance when using the proposed regularization scheme as compared to constant regularization scheme.

• Journal of the Computational Structural Engineering Institute of Korea
• /
• v.17 no.1
• /
• pp.1-10
• /
• 2004

Adaptive finite element analysis, in which its solution error meets with the user defined allowable error, is recently used to improve the reliability of finite element analysis results. This adaptive analysis is composed of two procedures; one is the error estimation of an analysis result and the other is the reconstruction of finite elements. In the (p-method, an element size is controlled by relocating of nodal positions (r-method) and the order of an element shape function is determined by the hierarchical polynomial (p-method) corresponding to the clement solution error by the enhanced SPR. In order to show the effectiveness and the accuracy of the suggested rp-method, various numerical examples were analyzed and these analysis results were examined by comparing with those obtained by the existed methods.

• Proceedings of the KSME Conference
• /
• 2000.04a
• /
• pp.345-350
• /
• 2000

Adaptive finite element analysis, which its solution error meets with the user defined allowable error, is recently used far improving reliability of finite element analysis results. This adaptive analysis is composed of two procedures; one is the error estimation of an analysis result and another is the reconstruction of finite elements. In the rp-method, an element size is controlled by relocating of nodal positions(r-method) and the order of an element shape function is determined by the hierarchical polynomial(p-method) corresponding to the element solution error. In order to show the effectiveness and accuracy of the suggested rp-method, various numerical examples were analyzed and these analysis results were examined by comparing with those obtained by the existed methods. As a result of this study, following conclusions are obtained. (1) rp-method is more accurate and effective than the r- and p-method. (2) The solution convergency of the rp-method is controlled by means of the iterative calculation numbers of the r- and p- method each other.

• Journal of Korea Multimedia Society
• /
• v.14 no.8
• /
• pp.1000-1009
• /
• 2011

Image interpolation is a technique which estimates the non-allocated pixel values on image scale-transform. It requires minimum computational complexity and minimum image quality degradation on the interpolated resultant images. In this paper we propose an image interpolation method using iterative error estimation. The proposed method consists of the following five steps: loss-information computational step, loss-information estimation step, loss-information application step, error computation step, and error application step. The experimental results obtained show that the average PSNR is increased by 3.3dB, subjective image quality is enhanced and the minimum computation complexity is decreased by 83%. The proposed image interpolation algorithm may be helpful in various applications such as image reconstruction and enlargement.

• Journal of Biomedical Engineering Research
• /
• v.30 no.1
• /
• pp.56-65
• /
• 2009

Compton imaging is often recognized as a potentially more valuable 3-D technique in nuclear medicine than conventional emission tomography. Due to inherent computational limitations, however, it has been of a difficult problem to reconstruct images with good accuracy. In this work we show that the row-action maximum likelihood algorithm (RAMLA), which have proven useful for conventional tomographic reconstruction, can also be applied to the problem of 3-D reconstruction of cone-beam projections from Compton scattered data. The major advantage of RAMLA is that it converges to a true maximum likelihood solution at an order of magnitude faster than the standard expectation maximiation (EM) algorithm. For our simulations, we first model a Compton camera system consisting of the three pairs of scatterer and absorber detectors placed at x-, y- and z-axes, and generate conical projection data using a software phantom. We then compare the quantitative performance of RAMLA and EM reconstructions in terms of the percentage error. The net conclusion based on our experimental results is that the RAMLA applied to Compton camera reconstruction significantly outperforms the EM algorithm in convergence rate; while computational costs of one iteration of RAMLA and EM are about the same, one iteration of RAMLA performs as well as 128 iterations of EM.

• The KIPS Transactions:PartB
• /
• v.12B no.5 s.101
• /
• pp.567-576
• /
• 2005

This paper proposes an image-based modeling method which recovers 3D models using projected line segments in multiple images. Using the method, a user obtains accurate 3D model data via several steps of simple manual works. The embedded nonlinear minimization technique in the model parameter estimation stage is based on the distances between the user provided image line segments and the projected line segments of primitives. We define an error using a finite line segment and thus increase accuracy in the model parameter estimation. The error is defined as the sum of differences between the observed image line segments provided by the user and the predicted image line segments which are computed using the current model parameters and camera parameters. The method is robust in a sense that it recovers 3D structures even from partially occluded objects and it does not be seriously affected by small measurement errors in the reconstruction process. This paper also describesexperimental results from real images and difficulties and tricks that are found while implementing the image-based modeler.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.252-255
• /
• 2020

Unsupervised deep learning methods have shown impressive results for the challenging monocular depth estimation task, a field of study that has gained attention in recent years. A common approach for this task is to train a deep convolutional neural network (DCNN) via an image synthesis sub-task, where additional views are utilized during training to minimize a photometric reconstruction error. Previous unsupervised depth estimation networks are trained within a fixed depth estimation range, irrespective of its possible range for a given image, leading to suboptimal estimates. To overcome this suboptimal limitation, we first propose an unsupervised adaptive depth estimation method guided by minimum and maximum (min-max) depth priors for a given input image. The incorporation of min-max depth priors can drastically reduce the depth estimation complexity and produce depth estimates with higher accuracy. Moreover, we propose a novel network architecture for adaptive depth estimation, called the AdaMM-DepthNet, which adopts the min-max depth estimation in its front side. Intensive experimental results demonstrate that the adaptive depth estimation can significantly boost up the accuracy with a fewer number of parameters over the conventional approaches with a fixed minimum and maximum depth range.