• The Journal of Korean Institute of Information Technology
• /
• v.17 no.5
• /
• pp.77-82
• /
• 2019

Diabetic retinopathy is a complicated form of diabetes due to circulatory disorder in the peripheral blood vessels of the retina. We segment the microvessel for diagnosing diabetic retinophathy. The conventional methods using filter and features can segment the thick blood vessels, but it has relatively weak for segmenting fine blood vessels. In pre-processing step, noise reduction filter and histogram equalization are applied to suppress the noise and enhance the image contrast. Then, deep learning technique is used for pixel-by-pixel segmentation. The accuracy of conventional methods is between 90% to 94%, while the proposed method has improved as 95% accuracy. There is a problem of segmentation error around the optic disc and exudate due to the network depth. However the accuracy can be improved by modifying the network architecture in the future.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2010.11a
• /
• pp.705-708
• /
• 2010

An eigen color co-occurrence approach is proposed that exploits the correlation between color channels to identify the degree of image similarity. This method is based on traditional co-occurrence matrix method and histogram equalization. On the purpose of feature extraction, eigen color co-occurrence matrices are computed for extracting the statistical relationships embedded in color images by applying Principal Component Analysis (PCA) on a set of color co-occurrence matrices, which are computed on the histogram equalized images. That eigen space is created with a set of orthogonal axes to gain the essential structures of color co-occurrence matrices, which is used to identify the degree of similarity to classify an input image to be tested for various purposes. In this paper RGB, Gaussian color space are compared with grayscale image in terms of PCA eigen features embedded in histogram equalized co-occurrence features. The experimental results are presented.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.05a
• /
• pp.59-62
• /
• 2021

본 논문에서는 사다리꼴 형태의 소속 함수를 적용하여 스트레칭 하는 과정에서 상한과 하한을 FCM 기반 양자화 기법을 적용하여 동적으로 조정하는 퍼지 스트레칭 기법을 제안한다. 제안된 퍼지 스트레칭 기법은 FCM 기반 양자화 기법을 적용하여 각 클러스터를 생성하고 생성된 각 클러스터의 중심에 해당되는 명암도를 이용하여 사다리꼴 형태의 소속 함수의 구간을 설정한다. 그리고 설정한 구간 정보를 이용하여 스트레칭을 위한 상한과 하한을 구하여 영상을 스트레칭 한다. 제안된 FCM 양자화 기반 퍼지 스트레칭 기법의 성능을 분석하기 위해서 명암도 분포가 좁고 명암 대비가 낮은 결절종 초음파 영상과 컨테이너 영상을 대상으로 실험하였다. 실험 결과에서도 알 수 있듯이 기존의 히스토그램 스트레칭 기법과 삼각형 형태의 소속 함수를 적용한 퍼지 스트레칭 기법보다 명암 대비가 향상되었다. 결절종 초음파 영상에서는 결절종 영역과 그 외의 영역 간의 명암 대비가 뚜렷하게 나타나서 결절종 추출에 효과적인 것을 확인하였고 컨테이너 영상에서는 컨테이너 데미지를 추출하는데 필요한 컨테이너 굴곡선 등과 같은 특징이 다른 기법들에 비해 선명하게 나타났다.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.10a
• /
• pp.186-188
• /
• 2021

본 논문에서는 초음파 영상에서 DBSCAN(Density-based spatial clustering of applications with noise)과 FCM 클러스터링 기반 양자화 기법을 적용하여 결절종을 추출하는 방법을 제안한다. 본 논문에서는 초음파 영상 촬영 시 좌우 상단의 지방층 영역과 하단 영역의 명암도가 어두운 영역을 잡음 영역으로 설정한다. 그리고 초음파 영상에 퍼지스트레칭 기법을 적용하여 잡음 영역을 최대한 제거 한 후에 ROI 영역을 추출한다. 추출된 ROI 영역에서 밀도 분포를 분석하기 위하여 히스토그램을 분석한 후에 DBSCAN을 적용하여 초음파 영상에서 결절종 후보에 해당되는 명암도를 추출한다. 추출한 후보 명암도를 대상으로 FCM 클러스터링 기법을 적용한다. FCM을 적용하는 단계에서 결절종의 저에코 혹은 무에코의 특징을 이용하여 클러스터 중심 값이 가장 낮은 클러스터를 양자화 한 후에 라벨링 기법을 적용시켜 결절종의 후보 객체를 추출한다. 제안된 결절종 추출 방법의 성능을 분석하기 위해 전문의가 결절종 영역을 표기한 초음파 영상과 표기되지 않은 초음파 영상 120쌍을 대상으로 DBSCAN, FCM, 그리고 제안된 방법 간의 성능을 비교 분석하였다. 제안된 방법에서는 120개의 초음파 영상에서 106개 결절종 영역이 추출되었고 FCM 기법에서는 80개가 추출되었고 DBSCAN에서는 36개가 추출되었다. 따라서 제안된 방법이 결절종 추출에 효율적인 것을 확인하였다.

• Journal of Digital Convergence
• /
• v.11 no.2
• /
• pp.299-307
• /
• 2013

The aim of this paper is the improvement of identification performance and the reduction of computational quantities in the face recognition system based on SIFT-Grid. Firstly, we propose a composition method of integrated template by removing similar SIFT keypoints and blending different keypoints in variety training images of one face class. The integrated template is made up of computation of similarity matrix and threshold-based histogram from keypoints in a same sub-region which divided by applying SIFT-Grid of training images. Secondly, we propose a computation method of similarity for identify of test image from composed integrated templates efficiently. The computation of similarity is performed that a test image to compare one-on-one with the integrated template of each face class. Then, a similarity score and a threshold-voting score calculates according to each sub-region. In the experimental results of face recognition tasks, the proposed methods is founded to be more accurate than both two other methods based on SIFT-Grid, also the computational quantities are reduce.

• Journal of Broadcast Engineering
• /
• v.19 no.2
• /
• pp.166-173
• /
• 2014

Recently, human action recognition have been developed for various broadcasting and video process. Since a video can consist of various scenes, keypoint approaches have been more attracted than template based methods for real application. Keypoint approahces tried to find regions having motion in video, and made 3-dimensional patches. Then, descriptors using histograms were computed from the patches, and a classifier based on machine learning method was applied to detect actions in video. However, a single classifier was difficult to handle various human actions. In order to improve this problem, approaches using multi classifiers were used to detect and to recognize objects. Thus, we propose a new human action recognition using decision-level fusion with support vector machine and sparse representation. The proposed method extracted descriptors based on keypoint approach from a video, and acquired results from each classifier for human action recognition. Then, we applied weights which were acquired by training stage to fuse each results from two classifiers. The experiment results in this paper show better result than a previous fusion method.

• Journal of the Korea Computer Graphics Society
• /
• v.17 no.4
• /
• pp.11-21
• /
• 2011

This paper proposes a method for decomposing a Hangul glyph of outline fonts into its initial, medial and final components using statistical-structural information. In a font family, the positions of components are statistically consistent and the stroke relationships of a Hangul character reflect its structure. First, we create the component histograms that accumulate the shapes and positions of the same components. Second, we make pixel clusters from character image based on pixel direction probabilities and extract the candidate strokes using position, direction, size of clusters and adjacencies between clusters. Finally, we find the best structural match between candidate strokes and predefined character model by relaxation labeling. The proposed method in this paper can be used for a study on formative characteristics of Hangul font, and for a font classification/retrieval system.

• Journal of Korea Multimedia Society
• /
• v.5 no.3
• /
• pp.255-262
• /
• 2002

A boundary sequence can be a good representation of arbitrary shapes, because it can represent them simply and precisely. However, boundary sequences have not been used as a representation of arbitrary shapes, because the pixel-based shape-features such as area, centroid, orientation, projection and so forth, could not be computed directly from them. In this paper, we show that the shape-features can be easily computed from the boundary sequences by introducing the cross-sections that are defined as vertical (or horizontal) line segments in a shape. A cross-section generation method is proposed, which generates cross-sections of the shape efficiently by tracing the boundary sequence of the shape once. Furthermore, a boundary sequence extraction method is also proposed, which generates a boundary sequence for each shape in a binary image automatically The proposed methods work well even if a shape has holes. Eventually, we show that a boundary sequence can be used effectively for representing arbitrary shapes.

• Journal of Broadcast Engineering
• /
• v.21 no.4
• /
• pp.562-568
• /
• 2016

In this paper, we focus on recognition of speed-limit signs among a few types of traffic signs because speed-limit sign is closely related to safe driving of drivers. Although histogram of oriented gradient (HOG) and local binary patterns (LBP) are representative features for object recognition, these features have a weakness with respect to rotation, in that it does not consider the rotation of the target object when generating patterns. Therefore, this paper propose the fast rotation-invariant binary patterns (FRIBP) algorithm to generate a binary pattern that is robust against rotation. The proposed FRIBP algorithm deletes an unused layer of the histogram, and eliminates the shift and comparison operations in order to quickly extract the desired feature. The proposed FRIBP algorithm is successfully applied to German Traffic Sign Recognition Benchmark (GTSRB) datasets, and the results show that the recognition capabilities of the proposed method are similar to those of other methods. Moreover, its recognition speed is considerably enhanced than related works as approximately 0.47second for 12,630 test data.

• Journal of KIISE
• /
• v.44 no.7
• /
• pp.692-702
• /
• 2017

Face detection using the LBP based feature descriptor has issues in that it can not represent spatial information between facial shape and facial components such as eyes, nose and mouth. To address these issues, in previous research, a facial image was divided into a number of square sub-regions. However, since the sub-regions are divided into different numbers and sizes, the division criteria of the sub-region suitable for the database used in the experiment is ambiguous, the dimension of the LBP histogram increases in proportion to the number of sub-regions and as the number of sub-regions increases, the sensitivity to facial orientation rotation increases significantly. In this paper, we present a novel facial region segmentation method that can solve in-plane rotation issues associated with LBP based feature descriptors and the number of dimensions of feature descriptors. As a result, the proposed method showed detection accuracy of 99.0278% from a single facial image rotated in orientation.