• The Journal of Korean Institute of Information Technology
• /
• v.17 no.5
• /
• pp.13-20
• /
• 2019

Recently, object detection technology based on CNN has been actively studied. Object detection technology is used as an important technology in autonomous vehicles, intelligent image analysis, and so on. In this paper, we propose a rotation change robust object detection system by applying MI-FL (Moment Invariant-Feature Layer) to SSD (Single Shot Multibox Detector) which is one of CNN-based object detectors. First, the features of the input image are extracted based on the VGG network. Then, a total of six feature layers are applied to generate bounding boxes by predicting the location and type of object. We then use the NMS algorithm to get the bounding box that is the most likely object. Once an object bounding box has been determined, the invariant moment feature of the corresponding region is extracted using MI-FL, and stored and learned in advance. In the detection process, it is possible to detect the rotated image more robust than the conventional method by using the previously stored moment invariant feature information. The performance improvement of about 4 ~ 5% was confirmed by comparing SSD with existing SSD and MI-FL.

• Proceedings of the Korea Multimedia Society Conference
• /
• 2004.05a
• /
• pp.656-659
• /
• 2004

최근 많은 연구에서, 동일한 영상그룹들로부터 추출된 저수준의 특징들을 이용해서 고수준의 정보를 분석한 뒤, 이를 이용해서 영상을 분류하는 방법들을 소개하고 있다. 이러한 연구는 CBIR의 인덱싱에서 저수준의 특징만을 사용할 때 발생하는 의미적인 차이(semantic gap)문제를 해결하여, 검색의 효율을 높일 수 있게 한다. 하지만 이들 연구는 대부분 전경(scenery)영상만을 대상으로 하고 있다. 한편 영상을 객체 단위로 다루는 것은 CBIR의 성능을 크게 향상 시킬 수 있는 요인이 된다. 왜냐하면 대부분의 사용자는 관심있는 객체가 포함된 영상을 검색하기 원하기 때문이다. 본 논문에서는 영상의 객체를 인공객체와 자연객체로 분류하는 방법을 제안한다. 인공객체의 경우 자연객체에 비해 상대적으로 직선형태의 에지가 많이 발견되며 객체를 구성하는 패턴이 규칙적이고 방향성을 가진다. 또한 인공객체는 자연객체에 비해 객체영역의 경계가 직선에 의한 단순한 형태로 나타난다. 이러한 특징들을 EDH(edge Direction Histogram)의 에너지, EDAS(Energy Difference of Adjacent Sector)와 가버 필터를 통해 추출하여 분류에 이용한다. 실험을 통하여 각 특징들을 개별적으로 사용해서 76%에서 84% 사이의 분류 정확성을 얻었으며, 제안한 머징 방법을 이용하여 최종적으로 약 90%의 정확성으로 분류하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2015.11a
• /
• pp.63-66
• /
• 2015

3차원 영상을 생성하기 위해 스테레오 매칭을 통해 깊이 정보를 획득한다. 이때 발생하는 경계영역과 텍스처가 부족한 부분의 깊이정보 부정확성 문제를 해결하기 위해 영상 분할 기반 스테레오 매칭 방법을 제안한다. 일반적으로 사용하는 윈도우 기반 스테레오 매칭 결과를 기반으로 분할된 영상 내에서 최적의 변위 값을 재 할당함으로서 깊이정보의 정확성을 향상시킬 수 있다. Mean-shift는 참조 영상에서 화소 간 평균값 차이가 최대가 되는 영역들을 반복적으로 찾는다. 유사한 평균값을 갖는 영역들을 기반으로 영상을 분할하는 것을 Mean-shift를 이용한 영상분할 이라고 한다. 분할된 영상은 각 영역을 대표하는 패치 구조를 가지고 있어 참조 영상에 포함되어있는 잡음에 강인한 특성을 지닌다. 스테레오 매칭을 통해 화소별로 변위 값을 할당해주는 대신, 분할된 영상을 이용하여 각 분할 영역에 동일한 변위 값을 할당한다. 분할된 영상에 동일한 변위 정보를 할당할 경우 객체와 배경의 경계영역에서 잘못된 변위 값이 할당되는 경우가 발생한다. 이러한 경계 영역의 변위정보 부정확성을 보완하기 위해 화소의 기울기 항을 비용 값 계산 과정에 추가하여 단점을 보완한다. 최종 비용 값 계산을 통해 획득한 초기 변위 지도에 중간 값 필터를 적용하여 분류된 영역에 동일한 변위 값을 할당한다. 제안한 방법을 적용하여 경계영역의 정확도가 향상된 최종 변위 지도를 획득한다.

• Journal of the Korea Convergence Society
• /
• v.6 no.6
• /
• pp.1-8
• /
• 2015

In this paper, we proposed image enhancement method using sharpening filter for improving the accuracy of object detection using the existing Grabcut algorithm. GrabCut algorithm is the excellent performance extracting an object within a rectangular window range, but it has the drawback of the inferior performance in image with no clear distinction between background and objects. So, in this paper, reinforcing the brightness and clarity through histogram equalization, and tightening the border of the object using the sharpening filter look better than that extracted result of existing GrabCut algorithm in a similar image of the object and the background. Based on improved Grabcut algorithm, it is possible to obtain an improved result in the image processing convergence technique of character recognition, real-time object tracking and so on.

• Journal of the Korea Society of Computer and Information
• /
• v.14 no.6
• /
• pp.59-64
• /
• 2009

Edge detection always influenced by the noise of original image, therefore need some methods to eliminate them in advance, and the Mean Shift algorithm has the smooth function which suit for this purpose, so adopt it to fade out the unimportant information and the sensitive noise portions. Above all, we use the Canny algorithm to pick up the contour of the objects we focus on. And, take tests and get better result than the former sole Canny algorithm. This combination method of Mean Shift algorithm and Canny algorithm is suitable for the edge detection processing.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2015.07a
• /
• pp.542-543
• /
• 2015

본 논문에서는 차량에 설치된 스테레오 카메라를 이용하여 객체 후보를 검출하는 기법을 제안한다. 단일 카메라를 이용하면 색상 정보만을 획득할 수 있지만 스테레오 카메라를 이용하면 추가적으로 깊이 영상(depth image)을 획득할 수 있다. 깊이 정보를 이용하면 카메라의 주변 환경을 재구성 할 수 있으며, 객체 후보 검출의 성능을 향상 시킬 수 있다. 스테레오 매칭(Stereo matching)을 통해 획득한 깊이 영상을 열 단위 분석을 통해 객체로 판단되는 영역을 검출한다. 검출한 영역 경계 부근의 오류를 줄이기 위해 밝기 차를 분석한다. 실험을 통해 제안하는 시스템이 효과적으로 객체 후보를 검출하는 것을 확인하였다.

• Journal of the Korea Convergence Society
• /
• v.9 no.11
• /
• pp.61-68
• /
• 2018

In this paper, we propose a method of character detection by analyzing image enhancement and character type to detect characters in natural images that can be acquired in everyday life. The proposed method emphasizes the boundaries of the object part using the unsharp mask in order to improve the detection rate of the area to be recognized as a character in a natural image. By using the boundary of the enhanced object, the character candidate region of the image is detected using Maximal Stable Extermal Regions (MSER). In order to detect the region to be judged as a real character in the detected character candidate region, the shape of each region is analyzed and the non-character region other than the region having the character characteristic is removed to increase the detection rate of the actual character region. In order to compare the objective test of this paper, we compare the detection rate and the accuracy of the character region with the existing methods. Experimental results show that the proposed method improves the detection rate and accuracy of the character region over the existing character detection method.

• Journal of Advanced Navigation Technology
• /
• v.23 no.1
• /
• pp.77-83
• /
• 2019

In this paper, we propose an object tracking algorithm based on color histogram and convolutional neural network model. In order to increase the tracking accuracy, we synthesize generic object tracking using regression network algorithm which is one of the convolutional neural network model-based tracking algorithms and a mean-shift tracking algorithm which is a color histogram-based algorithm. Both algorithms are classified through support vector machine and designed to select an algorithm with higher tracking accuracy. The mean-shift tracking algorithm tends to move the bounding box to a large range when the object tracking fails, thus we improve the accuracy by limiting the movement distance of the bounding box. Also, we improve the performance by initializing the tracking start positions of the two algorithms based on the average brightness and the histogram similarity. As a result, the overall accuracy of the proposed algorithm is 1.6% better than the existing generic object tracking using regression network algorithm.

• Korean Journal of Remote Sensing
• /
• v.33 no.6_2
• /
• pp.1089-1100
• /
• 2017

This paper presented a method using past land cover maps in image segmentation and training set collection for updating land cover maps. In this method, the object boundaries in past land cover maps were used for segmenting image clearly. Also, the classes of past land cover maps were used to extract additional informative training set from the initial classification result using a small number of initial training set. To evaluate the applicability of proposed method, a case study for updating land cover maps was carried out using middle-level land cover maps and WorldView-2 image in the Taean-gun, South Korea. As a result of the case study, the confusions between urban and barren, paddy/dry field and grassland in the initial classification result were reduced by adding training set. In addition, the object segmentation using boundaries of past land cover map cleared land cover boundaries and improved classification accuracy. Based on the result of case study, the proposed method using past land cover maps is expected to be useful for updating land cover maps.

• Journal of Broadcast Engineering
• /
• v.13 no.6
• /
• pp.915-929
• /
• 2008

In this paper, we propose an algorithm for pixel-based disparity estimation with reliability in the multi-view image. The proposed method estimates an initial disparity map using edge information of an image, and the initial disparity map is used for reducing the search range to estimate the disparity efficiently. Furthermore, disparity-mismatch on object boundaries and textureless-regions get reduced by adaptive block size. We generated intermediate-view images to evaluate the estimated disparity. Test results show that the proposed algorithm obtained $0.1{\sim}1.2dB$ enhanced PSNR(peak signal to noise ratio) compared to conventional block-based and pixel-based disparity estimation methods.