• Journal of Internet Computing and Services
• /
• v.18 no.6
• /
• pp.25-34
• /
• 2017

Autoencoder is a type of deep learning method where input layer and output layer are the same, and effectively extracts and restores characteristics of input vector using constraints of hidden layer. In this paper, we propose methods of Autoencoders to remove a natural background image which is a noise to the CAPTCHA and recover only a numerical images by applying various autoencoder models to a region where one number of CAPTCHA images and a natural background are mixed. The suitability of the reconstructed image is verified by using the softmax function with the output of the autoencoder as an input. And also, we compared the proposed methods with the other method and showed that our methods are superior than others.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2013.10a
• /
• pp.779-781
• /
• 2013

As the demand for various multimedia service increases the technology that utilizes image as information transfer method develops rapidly. Though average filter, median filter and weight filter etc. have been proposed to remove various noises that are added to images, the existing methods are short of noise removal and edge reservation performance. Therefore, in this paper an algorithm, in which noise is decided at the first hand, and then it is processed through modified median filter and adaptive weighted average filter, is proposed to effectively remove the complex noise that has been added to an image. And it was compared with existing methods through simulation and PSNR(peak signal to noise ratio) has been used as a criterion.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.16 no.5
• /
• pp.1064-1070
• /
• 2012

Nowadays, image processing has been applied in a variety of fields. In order to preserve the high quality of visual the degradation phenomenon for images should be removed. Noise is one of the representative elements cause of the degradation phenomenon and AWGN(additive white Gaussian noise) always damages images. In this paper, an mixed filter algorithm, which is based on parallel denoising method, is proposed to suppress AWGN. This algorithm parallels the spatial domain wiener filter and the wavelet domain thresholding method which thresholding function is selected based on scale level. The proposed modified thresholding function which considers the dependency between parent and child coefficient performs well on suppressing noise.

• Journal of Electrical Engineering and Technology
• /
• v.13 no.3
• /
• pp.1383-1391
• /
• 2018

A new nonlinear filtering algorithm for effectively denoising images corrupted by the random-valued impulse noise, called dual sliding statistics switching median (DSSSM) filter is presented in this paper. The proposed DSSSM filter is made up of two subunits; i.e. Impulse noise detection and noise filtering. Initially, the impulse noise detection stage of DSSSM algorithm begins by processing the statistics of a localized detection window in sorted order and non-sorted order, simultaneously. Next, the median of absolute difference (MAD) obtained from both sorted statistics and non-sorted statistics will be further processed in order to classify any possible noise pixels. Subsequently, the filtering stage will replace the detected noise pixels with the estimated median value of the surrounding pixels. In addition, fuzzy based local information is used in the filtering stage to help the filter preserves the edges and details. Extensive simulations results conducted on gray scale images indicate that the DSSSM filter performs significantly better than a number of well-known impulse noise filters existing in literature in terms of noise suppression and detail preservation; with as much as 30% impulse noise corruption rate. Finally, this DSSSM filter is algorithmically simple and suitable to be implemented for electronic imaging products.

• Journal of Magnetics
• /
• v.21 no.4
• /
• pp.593-598
• /
• 2016

In order to reduce the amount of noise component in X-ray imaging system, various reduction techniques were frequently used in the field of diagnostic imaging. Although the previous techniques -such as median, Wiener filters and Anscombe noise reduction technique - were able to reduce the noise, the edge information was still damaged. In order to cope with this problem, total variation (TV) noise reduction technique has been developed and researched. The purpose of this study was to evaluate and compare the image quality using normalized noise power spectrum (NNPS) and contrast-to-noise ratio (CNR) through simulations and experiments with respect to the above-mentioned noise reduction techniques. As a result, not only lowest NNPS value but also highest CNR values were acquired using a TV noise reduction technique. In conclusion, the results demonstrated that TV noise reduction technique is proved as the most practical method to ensure accurate denoising in X-ray imaging system.

• International journal of advanced smart convergence
• /
• v.8 no.1
• /
• pp.141-146
• /
• 2019

These days, there are many applications using neural networks as parts of their system. On the other hand, adversarial examples have become an important issue concerining the security of neural networks. A classifier in neural networks can be fooled and make it miss-classified by adversarial examples. There are many research to encounter adversarial examples by using denoising methods. Some of them using GAN (Generative Adversarial Network) in order to remove adversarial noise from input images. By producing an image from generator network that is close enough to the original clean image, the adversarial examples effects can be reduced. However, there is a chance when adversarial noise can survive the approximation process because it is not like a normal noise. In this chance, we propose a research that utilizes high-level representation in the classifier by combining GAN network with a trained U-Net network. This approach focuses on minimizing the loss function on high representation terms, in order to minimize the difference between the high representation level of the clean data and the approximated output of the noisy data in the training dataset. Furthermore, the generated output is checked whether it shows minimum error compared to true label or not. U-Net network is trained with true label to make sure the generated output gives minimum error in the end. At last, the remaining adversarial noise that still exist after low-level approximation can be removed with the U-Net, because of the minimization on high representation terms.

• Proceedings of the KIEE Conference
• /
• 2008.04a
• /
• pp.89-90
• /
• 2008

AWGN(Addictive white gaussian noise)에 의해 영상은 자주 훼손되곤 한다. 최근 이를 복원하기위해 웨이블릿(Wavelet) 영역에서의 베이시안(Bayesian) 추정법이 연구되고 있다. 웨이블릿 변환된 영상 신호의 밀도 함수(pdf)는 표족한 첨두와 긴 꼬리(long-tail)를 갖는 경망이 있다. 이러한 사전 밀도 함수(a priori probability density function)를 상황에 적합하게 추정한다면 좋은 성능의 복원 결과를 얻을 수 있다. 빈번이 제안되는 릴도 함수로 가우시안(Gaussian) 분포 참수와 라플라스(Laplace) 분포 함수가 있다. 이들 각각의 모델은 훌륭히 변환 계수들을 모델링하며 나름대로의 장점을 나타낸다. 본 연구에서는 가우시안 분포와 라플라스(Laplace) 분포의 혼합 분포 모델을 밀도 함수로 제안하여, 이 들의 장점을 종합하였다. 이를 MAP(Maximum a Posteriori) 추정 방법에 적용하여 잡음을 제거 하였다. 그 결과 기존의 알고리즘에 비해 시각적인 면(Visual aspect), 수치적인 면(PSNR), 그리고 연산량(Complexity) 측면에서 망상된 결과를 얻었다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.11a
• /
• pp.213-216
• /
• 2019

영상 잡음 제거 알고리즘은 잡음으로 오염된 영상으로부터 잡음이 제거된 깨끗한 영상을 추정하여 복원하는 연구이다. 기존의 모델 기반 방법의 영상 잡음 제거 알고리즘은 영상을 복원하는 과정에서 최적화 문제를 풀어야 한다는 단점과 매개변수를 직접 선택을 해주어야 한다는 단점을 가진다. 본 논문에서는 딥러닝을 이용한 학습기반 방법의 영상 잡음 제거 연구를 소개한다. 먼저, 신경망의 구축을 위하여 신경망의 구성 요소는 Instance Normalization 과 컨볼루션 신경망을 이용한 모델을 제안하였고, 여러 연구 분야에서 좋은 성능을 보이는 U-Net 구조를 전체적인 구조로 차용하였다. 신경망의 학습을 위하여 DnCNN 에서 제안한 잡음을 학습하는 잔여 학습 기법을 채택하였고, 기존의 영상 잡음 제거 알고리즘의 단점인 결과 영상이 흐릿해지는 현상을 보완하기 위하여 생성적 적대 신경망 학습 방법을 적용하였다. 본 논문에서 제안한 신경망을 이용한 잡음 제거 영상의 결과가 기존의 연구 방법들 보다 인지적인 측면에서 좋은 결과를 보임을 확인하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2014.06a
• /
• pp.226-229
• /
• 2014

본 논문은 효과적인 열화영상의 복원을 위해 Multiresolution Bilateral Filter (MBF) 기반의 구간적 노이즈 분석을 제안한다. 기존의 MBF 알고리즘은 최적화되지 않은 노이즈 추정 값을 중첩적으로 사용하다보니 over smoothing 현상이 발생되는 결과가 도출되기도 하였다. 이에 따른 보완점으로 열화영상 내 전체 화소를 일정한 블록 단위의 영역으로 나누어, 영상특성을 최대한 보존하며 노이즈제거를 진행하기 위해 블록 단위의 영역 내에서 노이즈 추정을 위한 파라미터를 추가한다. 실험을 통해 제안된 알고리즘이 노이즈 추정을 수행하여 얻어진 노이즈의 분산 값을 보다 정확히 추정하였고, 이로 인하여 향상된 노이즈 제거 영상 획득이 가능함을 확인할 수 있었다.

• Proceedings of the Korea Society of Information Technology Applications Conference
• /
• 2007.05a
• /
• pp.9-28
• /
• 2007

It is well known that wavelet transform is a signal processing technique which can display the signals on in both time and frequency domain. In this paper, we proposed a new approach based on stationary wavelet transform to provide an enhanced approach for eliminating noise. A 'stationary wavelet transform', where the coefficient sequences are not decimated at each stage, is described. The testing result on sample iris images has shown an enhanced image quality and also show that it has a superior performance than traditional discrete wavelet transform.