• ETRI Journal
• /
• v.46 no.3
• /
• pp.513-525
• /
• 2024

Deep neural networks trained on labeled medical data face major challenges owing to the economic costs of data acquisition through expensive medical imaging devices, expert labor for data annotation, and large datasets to achieve optimal model performance. The heterogeneity of diseases, such as Alzheimer's disease, further complicates deep learning because the test cases may substantially differ from the training data, possibly increasing the rate of false positives. We propose a reconstruction-based self-supervised anomaly detection model to overcome these challenges. It has a dual-subnetwork encoder that enhances feature encoding augmented by skip connections to the decoder for improving the gradient flow. The novel encoder captures local and global features to improve image reconstruction. In addition, we introduce an entropy-based image conversion method. Extensive evaluations show that the proposed model outperforms benchmark models in anomaly detection and classification using an encoder. The supervised and unsupervised models show improved performances when trained with data preprocessed using the proposed image conversion method.

• Proceedings of the IEEK Conference
• /
• 2002.07a
• /
• pp.599-602
• /
• 2002

A new approach of error concealment using a digital watermarking technique for interframe video coding is presented in this paper. In the proposed method, the most important feature of the reference frame is extracted. Then, this feature is embedded into the prediction error of current frame prior to transmission. Error concealment is achieved by means of recovering the erroneous reference frame using the embedded data before the reconstruction of current frame is performed. Simulation results demonstrated the effectiveness of the proposed method.

• Proceedings of the KSRS Conference
• /
• 1999.11a
• /
• pp.304-309
• /
• 1999

In this paper, we presents a new disparity map refinement algorithm using statistical characteristics of disparity map and edge information. The proposed algorithm generate a refined disparity map using disparity maps which are obtained from area and feature-based Stereo Matching by selecting a disparity value of edge point based on the statistics of both disparity maps. Experimental results on aerial stereo image show the better results than conventional fusion algorithms in the disparity error. This algorithm can be applied to the reconstruction of building image from the high resolution remote sensing data.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.30 no.1 s.244
• /
• pp.8-15
• /
• 2006

A Sparse Point Representation(SPR) based on interpolation wavelets is presented. The SPR is implemented for the purpose of CFD data compression. Unlike conventional wavelet transformation, the SPR relieves computing workload in the similar fashion of lifting scheme that includes splitting and prediction procedures in sequence. However, SPR skips update procedure that is major part of lifting scheme. Data compression can be achieved by proper thresholding method. The advantage of the SPR method is that, by keeping even point physical values, low frequency filtering procedure is omitted and its related unphysical thresholing mechanism can be avoided in reconstruction process. Extra singular feature detection algorithm is implemented for preserving singular features such as shock and vortices. Several numerical tests show the adequacy of SPR for the CFD data. It is also shown that it can be easily extended to nonlinear adaptive wavelets for enhanced feature capturing.

• Proceedings of the IEEK Conference
• /
• 2000.11c
• /
• pp.113-116
• /
• 2000

Human face recognition and 3D human face reconstruction has been studied in this paper. To find the facial feature points, find edge from input image and analysis the accumulated histogram of edge information. This paper use a Generic Face Model to display the 3D human face model which was implement with OpenGL and generated with 500 polygons. For reality of 3D human face model, we propose Group matching mapping method between facial feature points and the one of Generic Face Model. The personalized 3D human face model which resembles real human face can be generated automatically in less than 5 seconds on Pentium PC.

• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.25-28
• /
• 2002

Voice activity detection is important Problem in the speech recognition and communication. This paper introduces feature parameter which is reconstructed by the spectral entropy of information theory for the robust voice activity detection in the noise environment, analyzes and compares it with the energy method of voice activity detection and performance. In experiment, we confirmed that the spectral entropy is more feature parameter than the energy method for the robust voice activity detection in the various noise environment.

• Proceedings of the Korean Information Science Society Conference
• /
• 2008.06a
• /
• pp.301-302
• /
• 2008

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.181.5-181
• /
• 2001

This paper presents the method of 3D reconstruction of the depth information from the endoscopic stereo scopic images. After camera modeling to find camera parameters, we performed feature-point based stereo matching to find depth information. Acquired some depth information is finally 3D reconstructed using the NURBS(Non Uniform Rational B-Spline) algorithm. The final result image is helpful for the understanding of depth information visually.

• Proceedings of the Korean Information Science Society Conference
• /
• 2000.04b
• /
• pp.583-585
• /
• 2000

본 논문에서는 적은 수의 특징점을 이용한 얼굴 복원 방법을 제안한다. 먼저 얼굴을 형태와 질갑 프로토타입들의 선형 중첩으로 모형화한 다음, 특징점에서의 형태와 질감정보만을 가지고 각각의 얼굴이 요구하는 변형의 근사값을 찾는다. 본 논문에서는 이러한 under-determined 조건에서 최소 제곱법(least square minimization method)을 사용하여 최적값을 얻는다. 실험을 통하여 적은 수의 특징점을 이용하여 2차원 얼굴 영상을 효율적으로 복원할 수 있음을 검증하였다. 우리는 제안된 얼굴 영상을 압축하거나 겹침이나 잡영에 의해 손상된 영상으로부터 원래의 전체 정보를 복원하는데 중요한 역할을 할 수 있을 것으로 기대한다.

• 제어로봇시스템학회:학술대회논문집
• /
• 2000.10a
• /
• pp.541-541
• /
• 2000

This paper Presents the method of 3D reconstruction of the depth information from the endoscopic stereo scopic images. After camera modeling to find camera parameters, we peformed feature-point based stereo matching to find depth information. Acquired some depth information is finally 3D reconstructed using the NURBS(Non Uniform Rational B-Spline) algorithm. The final result image is helpful for the understanding of depth information visually.