• Annual Conference of KIPS
• /
• 2019.10a
• /
• pp.1045-1048
• /
• 2019

최근 CCTV 사용이 보편화되면서 방범 목적으로 서비스 시설이나 공공시설에 설치되는 CCTV의 수가 급격하게 증가하고 있다. 그에 따라 CCTV를 감시하는 노동력이 부족해지는 문제가 발생하여 이를 대체하기 위해 카메라 영상을 통하여 한번 인식한 사람을 다른 시간이나 장소에서 촬영된 영상에서 다시 인식하는 사람 재인식 기술이 주목받고 있다. 또한, 이러한 사람 재인식 기술은 보안 분야뿐만 아니라 영화나 드라마와 같은 영상 컨텐츠에 적용되어 불법 복제물을 찾는 일에 사용될 수도 있다. 기존의 사람 재인식에는 이미지의 유사도를 계산하는 방법이 사용되었지만, 조명이나 카메라 각도가 달라지면 성능이 급격하게 떨어지는 문제가 있었다. 최근에는 딥러닝 기술이 발달하면서 전반적인 영상처리 분야의 성능이 향상되었고, 사람 재인식 분야 역시 딥러닝을 활용하면서 성능이 향상되었다. 하지만 딥러닝을 활용한 방법의 경우 보통 두 개의 이미지를 입력으로 사용하여 같은지 다른지를 판단하게 되므로 각 이미지의 공통점이나 차이점을 동시에 고려하기는 어려운 점이 있다. 본 논문에서는 이러한 점을 해결하기 위해 세 개의 사람 이미지를 입력으로 사용하여 특징을 추출하고, 특징 맵을 재구성하여 각 이미지의 차이점과 공통점을 동시에 고려하며 학습할 수 있는 모델을 제안한다.

• Proceedings of the Korean Information Science Society Conference
• /
• 1999.10b
• /
• pp.353-355
• /
• 1999

본 논문에서는 영상의 flexible subblock을 이용하여 영상내에 물체의 이동이나, 빛의 변화, 시각점(view-point)의 변화등에 덜 민감한 영상 검색을 방법을 제안한다. 특징 값으로는 Ohta 컬러 공간으로부터 1, 2, 3차 central 모멘트 값을 추출해 내고, 쌍직교 웨이블릿 변환을 통해 고주파 영역으로부터 수직-수평 방향 성분을 추출하여 인덱스화 시킴으로써 인덱스를 위한 저장 공간을 줄이고 계산 시간을 향상시킬 수 있었다. 아울러, 2개의 특징 값을 다단계(multi-step) K-NN 방법에 적용시킴으로서 사용자가 검색하고자 하는 가장 유사한 k 개의 영상만을 사용자에게 보여 주도록 설계하였다. 본 논문에서는 제안하는 알고리즘의 우수성을 증명하기 위해 RGB 색상 공간을 그대로 적용하여 실험한 결과를 비교해 보았다. 추가적으로, 영상의 전역적인 유사성뿐만 아니라, 각 블록의 독립적인 특징 값을 이용하여 특정 블록에 대한 검색 환경도 제공하여 보다 의미있는 검색 환경을 제공하고 있다.

• Journal of Broadcast Engineering
• /
• v.17 no.5
• /
• pp.808-816
• /
• 2012

Face recognition has been actively studied and developed in various fields. In recent years, interest point extraction algorithms mainly used for object recognition were being applied to face recognition. The SURF(Speeded Up Robust Features) algorithm was used in this paper which was one of typical interest point extraction algorithms. Generally, the interest points extracted from human faces are less distinctive than the interest points extracted from objects due to the similar shapes of human faces. Thus, the accuracy of the face recognition using SURF tends to be low. In order to improve it, we propose a face recognition algorithm which performs interest point extraction by SURF and the Gabor wavelet transform to extract descriptors from the interest points. In the result, the proposed method shows around 23% better recognition accuracy than SURF-based conventional methods.

• Journal of IKEEE
• /
• v.23 no.1
• /
• pp.207-213
• /
• 2019

In this paper, we propose a low-complexity feature detection algorithm for object tracking and present hardware architecture design and implementation results for real-time processing. The existing Shi-Tomasi algorithm shows good performance in object tracking applications, but has a high computational complexity. Therefore, we propose an efficient feature detection algorithm, which can reduce the operational complexity with the similar performance to Shi-Tomasi algorithm, and present its real-time implementation results. The proposed feature detector was implemented with 1,307 logic slices, 5 DSP 48s and 86.91Kbits memory with FPGA. In addition, it can support the real-time processing of 54fps at an operating frequency of 114MHz for $1920{\times}1080FHD$ images.

• Journal of KIISE:Software and Applications
• /
• v.31 no.1
• /
• pp.71-81
• /
• 2004

Minutiae-based fingerprint identification systems use minutiae points, which cannot completely characterize local ridge structures. Further, this method requires many methods for matching two fingerprint images containing different number of minutiae points. Therefore, to represent the fired length information for one fingerprint image, the filterbank-based method was proposed as an alternative to minutiae-based fingerprint representation. However, it has two shortcomings. One shortcoming is that similar feature vectors are extracted from the different fingerprints which have the same fingerprint type. Another shortcoming is that this method has overload to reduce the rotation error in the fingerprint image acquisition. In this paper, we propose the minutia-weighted feature vector extraction method that gives more weight in extracting feature value, if the region has minutiae points. Also, we Propose new fingerprint alignment method that uses the average local orientations around the reference point. These methods improve the fingerprint system's Performance and speed, respectively. Experimental results indicate that the proposed methods can reduce the FRR of the filterbank-based fingerprint matcher by approximately 0.524% at a FAR of 0.967%, and improve the matching performance by 5% in ERR. The system speed is over 1.28 times faster.

• Korean Journal of Ichthyology
• /
• v.25 no.4
• /
• pp.222-226
• /
• 2013

As Carangoides ferdau was previously reported based on its underwater photograph, morphological descriptions have been incomplete up to the presence in Korea. On the base of two samples collected at the coast of Jeju island, morphological characters of C. ferdau are described in detail. This species is characterized by having the forepart of second dorsal fin much prolonged, 7~8 transverse dark bands on body, and snout length almost equal to eye diameter. It is morphologically very similar to C. orthogroammus, but is easily distinguished in having transverse dark bands instead of yellow spot on the body of C. orthogroammus. Phylogenetic relationships based on the mitochondrial cytochrome b (1,141 base pairs) sequences shows that C. ferdau is closely related to C. orthogroammus, and C. dinema also has a sister group relationship with C. ablongus. Both genetic distances (p-distances) are 8.2%, respectively.

• Journal of KIISE:Software and Applications
• /
• v.32 no.7
• /
• pp.612-624
• /
• 2005

In this paper, we propose a surface-based registration using a gaussian weighted distance map for PET-CT brain image fusion. Our method is composed of three main steps: the extraction of feature points, the generation of gaussian weighted distance map, and the measure of similarities based on weight. First, we segment head using the inverse region growing and remove noise segmented with head using region growing-based labeling in PET and CT images, respectively. And then, we extract the feature points of the head using sharpening filter. Second, a gaussian weighted distance map is generated from the feature points in CT images. Thus it leads feature points to robustly converge on the optimal location in a large geometrical displacement. Third, weight-based cross-correlation searches for the optimal location using a gaussian weighted distance map of CT images corresponding to the feature points extracted from PET images. In our experiment, we generate software phantom dataset for evaluating accuracy and robustness of our method, and use clinical dataset for computation time and visual inspection. The accuracy test is performed by evaluating root-mean-square-error using arbitrary transformed software phantom dataset. The robustness test is evaluated whether weight-based cross-correlation achieves maximum at optimal location in software phantom dataset with a large geometrical displacement and noise. Experimental results showed that our method gives more accuracy and robust convergence than the conventional surface-based registration.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.17 no.3
• /
• pp.197-205
• /
• 1992

In this paper object recognition using neural network is studied. The recognition is accomplished by matching linear line segments which are formed by local features extracted from the curvature points. Since there is similarities among segments. The boundary of models is not distinct in feature space. Due to these indistinctness the ambiguity of recognition occurs, and the recognition rate becomes degraded according to the limitation of boundary decision capability of neural network for similar of features. Object recognition and to improve recognition rate. Local features are used to represent the object effectively. The validity of the object recognition system is demonstrated by experiments for the occluded and varied objects.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.2
• /
• pp.57-64
• /
• 2009

This paper presents the hardware implementation of the pyramidal KLT(Kanade-Lucas-Tomasi) feature tracker. Because of its high computational complexity, it is not easy to implement a real-time KLT feature tracker using general-purpose processors. A hardware implementation of the pyramidal KLT feature tracker using FPGA(Field Programmable Gate Array) is described in this paper with emphasis on 1) adaptive adjustment of threshold in feature extraction under diverse lighting conditions, and 2) modification of the tracking algorithm to accomodate parallel processing and to overcome memory constraints such as capacity and bandwidth limitation. The effectiveness of the implementation was evaluated over ones produced by its software implementation. The throughput of the FPGA-based tracker was 30 frames/sec for video images with size of $720{\times}480$.

• Korean Journal of Remote Sensing
• /
• v.38 no.6_1
• /
• pp.1257-1271
• /
• 2022

Recently, many Earth observation optical satellites have been developed, as their demands were increasing. Therefore, a rapid preprocessing of satellites became one of the most important problem for an active utilization of satellite images. Satellite image matching is a technique in which two images are transformed and represented in one specific coordinate system. This technique is used for aligning different bands or correcting of relative positions error between two satellite images. In this paper, we propose an automatic image matching method among satellite images with different ground sampling distances (GSDs). Our method is based on improved feature matching and robust estimation of transformation between satellite images. The proposed method consists of five processes: calculation of overlapping area, improved feature detection, feature matching, robust estimation of transformation, and image resampling. For feature detection, we extract overlapping areas and resample them to equalize their GSDs. For feature matching, we used Oriented FAST and rotated BRIEF (ORB) to improve matching performance. We performed image registration experiments with images KOMPSAT-3A and RapidEye. The performance verification of the proposed method was checked in qualitative and quantitative methods. The reprojection errors of image matching were in the range of 1.277 to 1.608 pixels accuracy with respect to the GSD of RapidEye images. Finally, we confirmed the possibility of satellite image matching with heterogeneous GSDs through the proposed method.