• Proceedings of the Korea Database Society Conference
• /
• 2000.11a
• /
• pp.358-363
• /
• 2000

본 논문에서는 디지털오디오 원시 데이터의 양에 따라 적당한 양의 오디오워터마크를 생성, 삽입하여 일정한 수준의 오디오데이터의 품질을 유지하도록 하는 적응적 워터마킹을 제안한다. 제안하는 알고리즘은 심리음향모델인 frequency masking과 Wavelet 변환의 개념을 적용한다. 저작권자 혹은 소유자의 데이터는 PN-sequence를 이용하여 생성된다. 워터마크 생성량의 조절은 특정한 모듈이 담당하게 되는데 이 모듈은 원시 데이터의 크기에 따라 워터마크의 적당한 양을 산출하여 오디오데이터의 품질을 유지하도록 한다.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.10
• /
• pp.157-165
• /
• 2021

Recently, studies using deep learning to analyze a large amount of text are being actively conducted. In particular, a pre-trained language model that applies the learning results of a large amount of text to the analysis of a specific domain text is attracting attention. Among various pre-trained language models, BERT(Bidirectional Encoder Representations from Transformers)-based model is the most widely used. Recently, research to improve the performance of analysis is being conducted through further pre-training using BERT's MLM(Masked Language Model). However, the traditional MLM has difficulties in clearly understands the meaning of sentences containing new words such as newly coined words. Therefore, in this study, we newly propose NTM(Newly coined words Target Masking), which performs masking only on new words. As a result of analyzing about 700,000 movie reviews of portal 'N' by applying the proposed methodology, it was confirmed that the proposed NTM showed superior performance in terms of accuracy of sensitivity analysis compared to the existing random masking.

• Proceedings of the Korean Information Science Society Conference
• /
• 2001.04a
• /
• pp.859-861
• /
• 2001

본 논문은 오디오의 masking특성을 적용한 심리음향 모델을 이용하여 오디오의 고음질을 보장하면서 잡음과 압축 등의 공격에 강한 오디오 워터마킹 방법을 제안한다. 제안하는 워터마킹 방법은 심리음향 모델에 의해 생산되는 masking thresholds와 원신호의 power spectral density의 각 주파수별 차이 에너지를 이용하여 시간도메인에서 워터마크를 삽입하는 방법으로 오디오의 품질을 유지할 수 있다. 워터마크로는 자기상관성이 강한 PN-시퀀스를 이용하여 강인한 워터마킹을 구현한다. 그리고 PN-시퀀스와 같은 이진 시퀀스 워터마크의 검출을 위한 유사도 측정식을 제안한다.

• The KIPS Transactions:PartB
• /
• v.10B no.5
• /
• pp.491-496
• /
• 2003

We present a new watermark design tool for digital images and digital videos that are based on human visual system (HVS) characteristics. In this tool, basic mechanisms (inhibitory and excitatory behaviour of cells) of HVS are used to determine image dependent upper bound values on watermark insertion. This allows us to insert maximai allowable transparent watermark, which in turn is extremely hard to attack with common image processing, Motion Picture Experts Group (MPEG) compression. As the number of details (e.g. edges) increases in an image, the HVS decrease its sensitivity to the details. In the same manner, as the number of motion increases in a video signal, the HVS decrease its sensitivity to the motions. We model this decreased sensitivity to the details and motions as an (motion) entropy masking. Entropy masking model can be efficiently used to increase the robustness of image and video watermarks. We have shown that our entropy-masking model provides watermark scheme with increased transparency and henceforth increased robustness.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.5
• /
• pp.761-773
• /
• 2023

This paper proposes a method to increase the power-analysis resistance of the neural network model's feedforward process by replacing the exponential-based activation function, used in the deep-learning field, with an approximated function especially at the multi-layer perceptron model. Due to its nature, the feedforward process of neural networks calculates secret weight and bias, which already trained, so it has risk of exposure of internal information by side-channel attacks. However, various functions are used as the activation function in neural network, so it's difficult to apply conventional side-channel countermeasure techniques, such as masking, to activation function(especially, to exponential-based activation functions). Therefore, this paper shows that even if an exponential-based activation function is replaced with approximated function of simple form, there is no fatal performance degradation of the model, and than suggests a power-analysis resistant feedforward neural network with exponential-based activation function, by masking approximated function and whole network.

• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.335-336
• /
• 2000

A psychoacoustic model based noise shaping method is proposed, where noise's presence with a host signal will not be perceptually noticeable. The derivation of imperceptible noise levels from the masking thresholds of the signal involves a deconvolution associated with the spreading function in the psychoacoustic model, which results in an ill-conditioned problem. In this paper, the problem is formulated as a constrained optimization, and it is demonstrated that the solution provides noise shaping where the noise excitation level conforms to the masking thresholds of the signal.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.05a
• /
• pp.104-107
• /
• 2022

autoencoder deep learning model has excellent ability to restore abnormal data to normal data, so it is not appropriate for anomaly detection. In addition, the Inpainting method, which is a method of restoring hidden data after masking (masking) a part of the data, has a problem in that the restoring ability is poor for noisy images. In this paper, we use a method of modifying and improving the MLP-Mixer model to mask the image at a certain ratio and to reconstruct the image by delivering compressed information of the masked image to the model. After constructing a model learned with normal data from the MVTec AD dataset, a reconstruction error was obtained by inputting normal and abnormal images, respectively, and anomaly detection was performed through this. As a result of the performance evaluation, it was found that the proposed method has superior anomaly detection performance compared to the existing method.

• Journal of Korea Multimedia Society
• /
• v.8 no.11
• /
• pp.1432-1438
• /
• 2005

A psychoacoustic model based noise shaping method is proposed. The method shapes the noise in the frequency domain such that its presence with a host signal will not be perceptually noticeable. The derivation of imperceptible noise levels from the masking thresholds of the signal involves deconvolution associated with the spreading function in the psychoacoustic model. It has been known as an ill-conditioned Problem. In this paper, a constrained optimization is applied such that the noise excitation level conforms to the masking thresholds of the signal. Thus, the noises embedded in the signal will not be perceived by human ear, and its performance is demonstrated experimentally.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.327-332
• /
• 2023

데이터의 품질과 다양성은 모델 성능에 지대한 영향을 끼친다. 본 연구에서는 Topic을 활용한 데이터 전처리와 BERT 기반 MLM, T5, Random Masking을 이용한 증강으로 데이터의 품질과 다양성을 높이고자 했으며, 이를 KoBERT 기반 에세이 자동 평가 모델에 적용했다. 데이터 전처리만 진행했을 때, Quadratic Weighted Kappa Score(QWK)를 기준으로 모델이 에세이의 모든 평가 항목에 대해 베이스라인보다 더욱 높은 일치도를 보였으며 평가항목별 일치도의 평균을 기준으로 0.5368029에서 0.5483064(+0.0115035)로 상승했다. 여기에 제안하는 증강 방식을 추가 할 경우 MLM, T5, Random Masking 모두 성능 향상 효과를 보였다. 특히, MLM 데이터 증강 방식을 추가로 적용하였을 때 최종적으로 0.5483064에서 0.55151645(+0.00321005)으로 상승해 가장 높은 일치도를 보였으며, 에세이 총점으로 QWK를 기준으로 성능을 평가하면 베이스라인 대비 0.4110809에서 0.4380132(+0.0269323)로의 성능 개선이 있었다.

• The Journal of Engineering Research
• /
• v.2 no.1
• /
• pp.57-62
• /
• 1997

Two psychoacoustic models used in MPEG-1 employ different masking patterns, different masking indexes, and different computational procedures. As a result, Model 1 is inferior to Model 2 due to its worst case approach in computing the SMR even though it determines tonality and masking levels accurately. In this study, we investigate the performances of psychoacoustic models when we modify the MPEG-1 audio coder for variable bitrates. Simulation results show that Model 2 has a gain of 30 kbps in the dual channel mode and 20 kbps in the joint stereo mode. It is generally known that the joint stereo mode has a gain in bitrate compare to the dual channel mode. For signals with frequent attacks, this gain becomes larger in Model 1 than in Model 2. This is due to the fact that Model 1 uses the worst case approach in computing the SMR to reduce pre-echo