• Annual Conference of KIPS
• /
• 2011.04a
• /
• pp.33-35
• /
• 2011

얼굴 검출 시스템에서 얼굴의 검출율과 처리 속도는 중요한 고려사항이며 이 두 가지는 서로 비례 관계에 있다. 현재 얼굴 검출 시스템은 소프트웨어 처리가 대부분이며 소프트웨어만으로는 HD(High Definition)의 영상 처리는 어렵다. 따라서 본 실시간 얼굴 검출 영상처리를 위한 전처리 과정 중에 하나인 영상 축소기를 설계하고 이를 구현 및 검증하였으며 소프트웨어로 작성된 알고리즘과 처리 속도를 비교하였다.

• Annual Conference of KIPS
• /
• 2022.11a
• /
• pp.589-591
• /
• 2022

흉부 X-ray 영상은 폐 질환을 진단하는 기본적인 도구로써 널리 사용되고 있다. 정확한 진단을 위해 흉부 X-ray 영상의 품질을 평가하는 과정을 거쳐야 하는데, 이 과정은 주관적인 기준에 따라 수 작업으로 이루어지기 때문에 많은 시간과 비용이 소요된다. 따라서 본 논문에서는 임상 현장에서 사용되는 흉부 X-ray 영상 화질 평가 가이드라인을 기반으로 인공음영, 포함범위, 환자자세, 흡기정도, 그리고 투과 상태의 5가지 품질 평가를 자동화하는 도구를 제안한다. 제안하는 도구는 품질 판단에 소요되는 시간과 비용을 줄여주고, 더 나아가 흉부 병변 진단을 위한 학습 모델 개발의 양질의 학습 데이터를 선별하는 전처리 과정에 활용될 수 있다.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2019.01a
• /
• pp.465-467
• /
• 2019

본 논문에서는 영상생성이 가능한 딥러닝 네트워크를 이용하여 조영증강 CT 영상을 획득하는 연구를 수행하였다. CT는 고해상도 영상을 바탕으로 환자의 질병 및 암 세포 진단에 사용되는 의료영상 기법 중 하나이다. 특히, 조영제를 투여한 다음 CT 영상을 획득되는 영상을 조영증강 CT 영상이라 한다. 조영증강된 CT 영상은 물질의 구성 성분의 영상대비를 강조하여 임상의로 하여금 진단 및 치료반응 평가의 정확성을 향상시켜준다. 하지많은 수의 환자들이 조영제 부작용을 갖기 때문에 이에 해당되는 환자의 경우 조영증강 CT 영상 획득이 불가능해진다. 따라서 본 연구에서는 조영증강 영상을 얻지 못하는 환자 및 일반 환자의 불필요한 방사선의 노출을 최소화 하기 위하여 영상생성 딥러닝 기법을 이용하여 CT 영상에서 조영증강 CT 영상을 생성하는 연구를 진행하였다. 영상생성 딥러닝 네트워크는 generative adversarial network (GAN) 모델을 사용하였다. 연구결과 아무런 전처리도 거치지 않은 CT 영상을 이용하여 영상을 생성하는 것 보다 히스토그램 균일화 과정을 거친 영상이 더 좋은 결과를 나타냈으며 생성영상이 기존의 실제 영상과 영상의 구조적 유사도가 높음을 확인할 수 있다. 본 연구결과 딥러닝 영상생성 모델을 이용하여 조영증강 CT 영상을 생성할 수 있었으며, 이를 통하여 환자의 불필요한 방사선 피폭을 최소하며, 생성된 조영증강 CT 영상을 바탕으로 정확한 진단 및 치료반응 평가에 기여할 수 있을거라 기대된다.

• Tunnel and Underground Space
• /
• v.34 no.3
• /
• pp.208-217
• /
• 2024

There is so much research effort for developing and implementing deep learning-based surveillance systems to manage health and safety issues in construction sites. Especially, the development of deep learning-based object detection in various environmental changes has been progressing because those affect decreasing searching performance of the model. Among the various environmental variables, the accuracy of the object detection model is significantly dropped under low illuminance, and consistent object detection accuracy cannot be secured even the model is trained using low-light images. Accordingly, there is a need of low-light enhancement to keep the performance under low illuminance. Therefore, this paper conducts a comparative study of various deep learning-based low-light image enhancement models (GLADNet, KinD, LLFlow, Zero-DCE) using the acquired construction site image data. The low-light enhanced image was visually verified, and it was quantitatively analyzed by adopting image quality evaluation metrics such as PSNR, SSIM, Delta-E. As a result of the experiment, the low-light image enhancement performance of GLADNet showed excellent results in quantitative and qualitative evaluation, and it was analyzed to be suitable as a low-light image enhancement model. If the low-light image enhancement technique is applied as an image preprocessing to the deep learning-based object detection model in the future, it is expected to secure consistent object detection performance in a low-light environment.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.19 no.11
• /
• pp.2106-2115
• /
• 1994

As a pre processing for an edge detection process. edge preserving smoothing algorithm is proposed. For this purpose we used the interpolation method using B-spline basis function and scaling of digital images. By approximation of continuous function from descrete data using B-spline basis function. undetermined data between two sample can be computed. so that they smooth the surfaces of objects. Some edges having mainly low frequency components are detected using down scaling of the images. Edge maps from proposed pre processed images are hardly affected by the varying space constants($\sigma$) and threshold values used in detecting zero-crossing.

• The Journal of the Korea Contents Association
• /
• v.8 no.6
• /
• pp.16-25
• /
• 2008

Today, mobile phone is a necessity of modern life. For that reason, we suggest a particular system of a mobile phone which take a picture of music score image and automatically play it without any technical knowledges about the music score information. This experiment makes midi, acknowleging separate symbols via preprocessing to music score image taken. This paper utilizes 11 sorts of the score image taken by a mobile phone camera for this experiment. Through this method we suggest, as much as 98% on average takes place, which is very high recognizing ratio. Also, as we introduce this system in a mobile phone by porting, it takes 8.63 seconds on average to create midi following input of images.

• Journal of Korean Society for Geospatial Information Science
• /
• v.1 no.1 s.1
• /
• pp.181-192
• /
• 1993

Image is composed of the digital numbers including information on natural phenomena, their condition and the kind of objects. Digital numbers change in geometric correction(that is preprocessing). This change of digital numbers gave an effect on results of land-cover classification. We intend to know the influence of resampling as classifying land-cover using the image reconstructed by geometric correction in this paper. Chun-cheon basin was selected the study area having most variable land-cover pattern in North-Han river valley and made on use of RESTEC data resampled in preprocessing. Land-cover is classified as six classes of LEVEL I using maximum likelyhood classification method. We classified land-cover using the image resampled by two methods in this study. Bilinear interpolation method was most accurate in five classes except bear-land in the result of comparing each class with topographic map. We should choose the method of resampling according to the class in which we put the importance in the image resampling of geometric correction. And if we use four-season's image, we may classify more accurately in case of the confusion in case of the confusion in borders of rice field and farm.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.37 no.2
• /
• pp.50-59
• /
• 2000

In this paper, a new preprocessing method is proposed to speedup the watershed-based image segmentation In the proposed method, the gradient correction values of ramp edges are calculated from the positions and width of the ramp edges using Laplacian operator, and then, unlike the conventional method in which the monoscale or multi scale gradient image is directly used as a reference iImage, the reference image is obtained by adding the threshold value to each position of the ramp edges in the monoscale gradient image And the marker image is reconstructed on the reference image by erosion By preprocessing the image for the watershed transformation in such a manner, we can reduce the oversegmentations far more than those of applying the conventional morphological filter to the simple monoscale or multiscale gradient-based reference image Thus, we can reduce the total image segmentation time by reducing the time of postprocessing of region merging, which consumes most of the processing time In the watershed-based image segmentation, Experimental results indicate that the proposed method can speedup the total image segmentation about twice than those of the conventional methods, without the loss of ramp edges and principal edges around the dense-edge region.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.36S no.11
• /
• pp.34-43
• /
• 1999

A high-speed image processor (HIP) is implemented for a high-speed multi-function peripheral. HIP has a binarization architecture with unified data path. It has the pixel-by-pixel pipelined processing to minimize size of the external memory. It performs pre-processing such as shading correction, automatic gain control (AGC), and gamma correction, and also drives external CCD or CIS modules. The pre-processed data can be enlarged or reduced. Various binarizatin algorithms can be processed in the unified archiecture. The embedded binarization algorithms are simple thresholding, high pass filtering, dithering, error diffusion, and thershold modulated error diffusion. These binarization algorithms are unified based on th threshold modulated error diffusion. The data path is designed to share the common functional block of the binarization algorithms. The complexity of the controls and the gate counts is greatly reduced with this novel architecture.

• Journal of Broadcast Engineering
• /
• v.13 no.3
• /
• pp.299-309
• /
• 2008

In this paper, we propose an algorithm for object generation from model-based 3-dimensional multi-viewpoint images using OpenGL rendering. In the first step, we preprocess a depth map image in order to get a three-dimensional coordinate which is sampled as a vertex information on OpenGL and has a z-value as depth information. Next, the Delaunay Triangulation algorithm is used to construct a polygon for texture-mapping using the vertex information. Finally, by mapping a texture image on the constructed polygon, we generate a viewpoint-adaptive object by calculating 3-dimensional coordinates on OpenGL.