• Journal of the Korea Society of Computer and Information
• /
• v.27 no.2
• /
• pp.33-42
• /
• 2022

In this paper, we proposed a deep learning model based on multi-scale dilated convolution feature fusion for the segmentation of black ice region in road image to send black ice warning to drivers in real time. In the proposed multi-scale dilated convolution feature fusion network, different dilated ratio convolutions are connected in parallel in the encoder blocks, and different dilated ratios are used in different resolution feature maps, and multi-layer feature information are fused together. The multi-scale dilated convolution feature fusion improves the performance by diversifying and expending the receptive field of the network and by preserving detailed space information and enhancing the effectiveness of diated convolutions. The performance of the proposed network model was gradually improved with the increase of the number of dilated convolution branch. The mIoU value of the proposed method is 96.46%, which was higher than the existing networks such as U-Net, FCN, PSPNet, ENet, LinkNet. The parameter was 1,858K, which was 6 times smaller than the existing LinkNet model. From the experimental results of Jetson Nano, the FPS of the proposed method was 3.63, which can realize segmentation of black ice field in real time.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.9
• /
• pp.1-12
• /
• 2022

Traditional methods of measuring corneal ulcers were difficult to present objective basis for diagnosis because of the subjective judgment of the medical staff through photographs taken with special equipment. In this paper, we propose a method to detect the ulcer area on a pixel basis in corneal ulcer images using a semantic segmentation model. In order to solve this problem, we performed the experiment to detect the ulcer area based on the DeepLab model which has the highest performance in semantic segmentation model. For the experiment, the training and test data were selected and the backbone network of DeepLab model which set as Xception and ResNet, respectively were evaluated and compared the performances. We used Dice similarity coefficient and IoU value as an indicator to evaluate the performances. Experimental results show that when 'crop & resized' images are added to the dataset, it segment the ulcer area with an average accuracy about 93% of Dice similarity coefficient on the DeepLab model with ResNet101 as the backbone network. This study shows that the semantic segmentation model used for object detection also has an ability to make significant results when classifying objects with irregular shapes such as corneal ulcers. Ultimately, we will perform the extension of datasets and experiment with adaptive learning methods through future studies so that they can be implemented in real medical diagnosis environment.

• The Journal of the Korea Contents Association
• /
• v.10 no.1
• /
• pp.1-9
• /
• 2010

In this paper we present a shading technique for realistic rendering of the surfaces with irregular and complex textures using a single photograph. So far, most works have been using many photographs or special photographing equipment to render the surfaces with irregular and complex textures as well as dividing texture regions manually. We present an automatic selection method of the region segmentation techniques according to properties of materials. As our technique produces a reflectance model and the approximated Bidirectional Reflection Distribution Function(BRDF) parameters, it allows the recovery of the photometric properties of diffuse, specular, isotropic or anisotropic textured objects. Also it make it possible to present several synthetic images with novel lighting conditions and views.

• Journal of Digital Convergence
• /
• v.11 no.7
• /
• pp.157-163
• /
• 2013

In this paper, we propose a vision and depth information based real-time hand gesture interface method using finger joint estimation. For this, the areas of left and right hands are segmented after mapping of the visual image and depth information image, and labeling and boundary noise removal is performed. Then, the centroid point and rotation angle of each hand area are calculated. Afterwards, a circle is expanded at following pattern from a centroid point of the hand to detect joint points and end points of the finger by obtaining the midway points of the hand boundary crossing and the hand model is recognized. Experimental results that our method enabled fingertip distinction and recognized various hand gestures fast and accurately. As a result of the experiment on various hand poses with the hidden fingers using both hands, the accuracy showed over 90% and the performance indicated over 25 fps. The proposed method can be used as a without contacts input interface in HCI control, education, and game applications.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.6
• /
• pp.813-826
• /
• 2022

AI-based image processing technologies in various fields have been widely studied. However, the lighter the board, the more difficult it is to reduce the weight of image processing algorithm due to a lot of computation. In this paper, we propose a method using deep learning for object recognition algorithm in lightweight embedded boards. We can determine the area using a deep neural network architecture algorithm that processes semantic segmentation with a relatively small amount of computation. After masking the area, by using more accurate deep learning algorithm we could operate object detection with improved accuracy for efficient neural network (ENet) and You Only Look Once (YOLO) toward executing object recognition in real time for lightweighted embedded boards. This research is expected to be used for autonomous driving applications, which have to be much lighter and cheaper than the existing approaches used for object recognition.

• The KIPS Transactions:PartD
• /
• v.14D no.4 s.114
• /
• pp.447-458
• /
• 2007

The objective of this study is to provide faster mobile web service by improving performance of Contents Cache used for mobile web service in the existing Mobile Gate System. It was found that two elements existed in Mark-Up page transcoded by Contents Generator. One of the elements was dependent only on the requested DIDL page and Mark-Up type. The other was dependent on each of the requested DIDL page, Mark-Up type, size of mobile display 모바일 장치 to request service, type of images available and color depth count of the images available. The conventional Contents Cache saved the entire Mark-Up page to hold both of the two elements. This caused the problem where storage space was not effectively used because reusable elements were repetitively saved in cache memory domain due to change in one of the elements even though all the other elements were the same. As a result, a larger number of transcoded Mark-Up pages could not be saved in the same cache memory size. Therefore, in this study, Mark-Up pages transcoded by Contents Generator were divided into two elements and were separately saved. Also, in order to respond to the demand for replacing data in cache with new data, this study applied two algorithms of LFU and LRU. This study proposed the method to implement cache performance of faster speed by enabling to save more number of the transcoded Mark-Up pages in the same cache storage space.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.39 no.3
• /
• pp.141-148
• /
• 2021

This paper presents a method to detect zebra-crossing using deep learning which combines SegNet and ResNet. For the blind, a safe crossing system is important to know exactly where the zebra-crossings are. Zebra-crossing detection by deep learning can be a good solution to this problem and robotic vision-based assistive technologies sprung up over the past few years, which focused on specific scene objects using monocular detectors. These traditional methods have achieved significant results with relatively long processing times, and enhanced the zebra-crossing perception to a large extent. However, running all detectors jointly incurs a long latency and becomes computationally prohibitive on wearable embedded systems. In this paper, we propose a model for fast and stable segmentation of zebra-crossing from captured images. The model is improved based on a combination of SegNet and ResNet and consists of three steps. First, the input image is subsampled to extract image features and the convolutional neural network of ResNet is modified to make it the new encoder. Second, through the SegNet original up-sampling network, the abstract features are restored to the original image size. Finally, the method classifies all pixels and calculates the accuracy of each pixel. The experimental results prove the efficiency of the modified semantic segmentation algorithm with a relatively high computing speed.

• Journal of Korea Multimedia Society
• /
• v.16 no.12
• /
• pp.1385-1392
• /
• 2013

There is a variety of researches about recognition system using biometric data these days. In this study, we propose a new algorithm, uses simultaneous equation that made of the edge of objects, to segment an iris region without threshold values from an anterior eye image. The algorithm attempts to find a center area through calculated outskirts information of an iris, and decides the area where the most points are accumulated. To verify the proposed algorithm, we conducted comparative experiments to Hough transform and Daugman's method, based on 50 images anterior eye images. It was found that proposed algorithm is 5 and 75 times faster than on each algorithm, and showed high accuracy of detecting a center point (95.36%) more than Hough transform (92.43%). In foreseeable future, this study is expected to useful application in diverse department of human's life, such as, identification system using an iris, diagnosis a disease using an anterior image.

• Proceedings of the Korea Inteligent Information System Society Conference
• /
• 2005.11a
• /
• pp.547-556
• /
• 2005

본 논문에서는 출입국자 관리의 효율성과 제계적인 출입국 관리를 위하여 여권 코드를 자동으로 인식하고 위조 여권을 판별할 수 있는 여권 인식 및 얼굴 인증 방법을 제안한다. 여권 이미지가 기울어진 상태로 스캔되어 획득되어질 경우 개별 코드 인식과 얼굴 인증에 많은 영향을 미칠 수도 있으므로 기울기 보정은 문자 분할 및 인식, 얼굴 인증에 있어 매우 중요하다. 따라서 본 논문에서는 여권 영상을 스미어링한 후, 추출된 문자열 중에서 가장 긴 문자열을 선택하고 이 문자열의 좌측과 우측 부분의 두께 중심을 연결하는 직선과 수평선과의 기울기를 이용하여 여권 영상에 대한 각도 보정을 수행한다. 여권 모드 추출은 소벨 연산자와 수평 스미어링, 8 방향 윤곽선 추적 알고리즘을 적용하여 여권 코드의 문자열 영역을 추출하고, 추출된 여권 코드 문자열 영역에 대해 반복 이지화 방법을 적용하여 코드의 문자열 영역을 이진화한다. 이진화된 문자열 영역에 대해 CDM 마스크를 적용하여 문자열의 코드들을 복원하고 8 방향 윤곽선 추적 알고리즘을 적용하여 개별 코드를 추출한다. 추출된 개별 코드 인식은 개선된 RBF 네트워크를 제안하여 적용한다. 제안된 RBF 네트워크는 퍼지 논리 접속 연산자를 이용하여 경계변수를 통적으로 조정하는 개선된 퍼지 ART 알고리즘을 제안하여 RBF 네트워크의 중간층으로 적용한다. 얼굴 인증을 위해서는 얼굴 인증에 가장 보편적으로 사용되는 PCA 알고리즘을 적용한다. PCA 알고리즘은 고차원의 벡터를 저 차원의 벡터로 감량하여 전체 입력 영상들의 직교적인 공분산행렬을 계산한 후 그것의 고유 값에 따라 각 영상의 고유벡터를 구하므로 PCA 알고리즘을 적용하여 얼굴의 고유 벡터를 구한 후 특징 벡터를 추출한다. 따라서 여권 영상에서 획득되어진 얼굴 영상의 특징벡터와 데이터베이스에 있는 얼굴 영상의 특징벡터와의 거리 값을 계산하여 사진 위조 여부를 판별한다. 제안된 여권 인식 및 얼굴 인증 방법의 성능을 평가를 위하여 원본 여권에서 얼굴 부분을 위조한 여권과 기울어진 여권 영상을 대상으로 실험한 결과, 제안된 방법이 여권의 코드 인식 및 얼굴 인증에 있어서 우수한 성능이 있음을 확인하였다.

• Proceedings of the Korean Institute of Navigation and Port Research Conference
• /
• 2004.04a
• /
• pp.265-268
• /
• 2004

At these days, electronic maritime chart is in positively using various fields. But most of electronic maritime chart and topography information system is hard to using various fields despite that demand because that difficult approach from separately and specific constitution. This thesis apply to image filtering that extract requirement topography data from simplification of the whole system. And as constructing database of simple structure that getting various information from sounding depth and existing database, propose electronic maritime chart system to practicable 3D embodiment. And also propose convenience and commonplace to simple process about frequently renewal.