• Journal of Korean Society for Geospatial Information Science
• /
• v.23 no.1
• /
• pp.81-88
• /
• 2015

Various satellite sensors having ranges of the visible, infrared, and thermal wavelengths have been launched due to the improvement of hardware technologies of satellite sensors development. According to the development of satellite sensors with various wavelength ranges, the fusion and integration of multisensor images are proceeded. Image matching process is an essential step for the application of multisensor images. Some algorithms, such as SIFT and SURF, have been proposed to co-register satellite images. However, when the existing algorithms are applied to extract matching points between optical and thermal images, high accuracy of co-registration might not be guaranteed because these images have difference spectral and spatial characteristics. In this paper, location of control points in a reference image is extracted by SURF, and then, location of their corresponding pairs is estimated from the correlation of the local similarity. In the case of local similarity, phase correlation method, which is based on fourier transformation, is applied. In the experiments by simulated, Landsat-8, and ASTER datasets, the proposed algorithm could extract reliable matching points compared to the existing SURF-based method.

• The KIPS Transactions:PartB
• /
• v.14B no.7
• /
• pp.493-502
• /
• 2007

An on-line registration technique is presented to register multi-view range images for the 3D reconstruction of real objects. Using a range camera, we first acquire range images and photometric images continuously. In the range images, we divide object and background regions using a predefined threshold value. For the coarse registration of the range images, the centroid of the images are used. After refining the registration of range images using a projection-based technique, we use a modified KLT(Kanade-Lucas-Tomasi) tracker to match photometric features in the object images. Using the modified KLT tracker, we can track image features fast and accurately. If a range image fails to register, we acquire new range images and try to register them continuously until the registration process resumes. After enough range images are registered, they are integrated into a 3D model in offline step. Experimental results and error analysis show that the proposed method can be used to reconstruct 3D model very fast and accurately.

• The Journal of Korean Institute of Next Generation Computing
• /
• v.15 no.5
• /
• pp.64-74
• /
• 2019

Although a non-rigid registration has high demands in clinical practice, it has a high computational complexity and it is very difficult for ensuring the accuracy and robustness of registration. This study proposes a method of applying a non-rigid registration to 3D magnetic resonance images of brain in an unsupervised learning environment by using a deep-learning network. A feature vector between two images is produced through the network by receiving both images from two different patients as inputs and it transforms the target image to match the source image by creating a displacement vector field. The network is designed based on a U-Net shape so that feature vectors that consider all global and local differences between two images can be constructed when performing the registration. As a regularization term is added to a loss function, a transformation result similar to that of a real brain movement can be obtained after the application of trilinear interpolation. This method enables a non-rigid registration with a single-pass deformation by only receiving two arbitrary images as inputs through an unsupervised learning. Therefore, it can perform faster than other non-learning-based registration methods that require iterative optimization processes. Our experiment was performed with 3D magnetic resonance images of 50 human brains, and the measurement result of the dice similarity coefficient confirmed an approximately 16% similarity improvement by using our method after the registration. It also showed a similar performance compared with the non-learning-based method, with about 10,000 times speed increase. The proposed method can be used for non-rigid registration of various kinds of medical image data.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.8 no.7
• /
• pp.1546-1552
• /
• 2004

Electronic Ink is a stored data in the form of the handwritten text or the script without converting it into ASCII by handwritten recognition on the pen-based computers and Personal Digital Assistants(Pocket PC) for supporting natural and convenient data input. One of the most important issues is to search the electronic ink in order to use it. We proposed and implemented a script matching algorithm for the electronic ink. Proposed matching algorithm separated the input stroke into a set of primitive stroke using the curvature of the stroke curve. After determining the type of separated strokes, it produced a stroke feature vector. And then it calculated the distance between the stroke feature vector of input strokes and one of strokes in the database using the dynamic programming technique.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.8 no.6
• /
• pp.1449-1457
• /
• 2007

A panorama image is constructed by merging several overlapped images to a big one. There are two kinds of methods, feature based and direct method, in the construction. Feature based one has a merit of processing speed faster than direct one. But, it is difficult to be implemented under slower processing environments such as mobile device. This paper proposed high speed construction method of a panorama image. The algorithm extremely improved matching speed by reducing the number of matching parameters using scene shot guider, and additionally adapted local matching technique to reduce matching error caused by the fewer matching parameters. In the experiments, it was shown that the proposed method required about 0.078 second in processing time, about 17 times shorter than the feature based one, for 24-bit color images of $320{\times}240$ size.

• Korean Journal of Remote Sensing
• /
• v.30 no.6
• /
• pp.787-795
• /
• 2014

This paper presents an image registration using Triangulation-based Local Transformation (TLT) applied to the remaining matched points after elimination of the matched points with gross error. The corners extracted using geometric mean-based corner detector are matched using Pearson's correlation coefficient and then accepted as initial matched points only when they satisfy the Left-Right Consistency (LRC) check. We finally accept the remaining matched points whose RANdom SAmple Consensus (RANSAC)-based global transformation (RGT) errors are smaller than a predefined outlier threshold. After Delaunay triangulated irregular networks (TINs) are created using the final matched points on reference and sensed images, respectively, affine transformation is applied to every corresponding triangle and then all the inner pixels of the triangles on the sensed image are transformed to the reference image coordinate. The proposed algorithm was tested using KOMPSAT-2 images and the results showed higher image registration accuracy than the RANSAC-based global transformation.

• Journal of KIISE:Computer Systems and Theory
• /
• v.31 no.3_4
• /
• pp.224-238
• /
• 2004

Multi-modality image registration is a widely used image processing technique to obtain composite information from two different kinds of image sources. This study proposes an image registration method based on moment information and surface distance, which improves the previous surface-based registration method. The proposed method ensures stable registration results with low registration error without being subject to the initial position and direction of the object. In the preprocessing step, the surface points of the object are extracted, and then moment information is computed based on the surface points. Moment information is matched prior to fine registration based on the surface distance, in order to ensure stable registration results even when the initial positions and directions of the objects are very different. Moreover, surface comer sampling algorithm has been used in extracting representative surface points of the image to overcome the limits of the existed random sampling or systematic sampling methods. The proposed method has been applied to brain MRI(Magnetic Resonance Imaging) and PET(Positron Emission Tomography), and its accuracy and stability were verified through registration error ratio and visual inspection of the 2D/3D registration result images.

• Journal of the Korea Society of Computer and Information
• /
• v.17 no.6
• /
• pp.37-48
• /
• 2012

Image registration is a process to establish the spatial correspondence between the images of same scene, which are acquired at different view points, at different times, or by different sensors. In this paper, we propose an effective registration method for images acquired by multi-sensors, such as EO (electro-optic) and IR (infrared) sensors. Image registration is achieved by extracting features and finding the correspondence between features in each input images. In the recent research, the multi-sensor image registration method that finds corresponding features by exploiting NMI (Normalized Mutual Information) was proposed. Conventional NMI-based image registration methods assume that the statistical correlation between two images should be global, however images from EO and IR sensors often cannot satisfy this assumption. Therefore the registration performance of conventional method may not be sufficient for some practical applications because of the low accuracy of corresponding feature points. The proposed method improves the accuracy of corresponding feature points by combining the gradient orientation as spatial information along with NMI attributes and provides more accurate and robust registration performance. Representative experimental results prove the effectiveness of the proposed method.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2016.06a
• /
• pp.29-32
• /
• 2016

특징점 기반 건물인식 시스템에서는 강건한 특징점을 추출하는 것이 인식률 향상에 바로 직결되는 중요한 요소이다. 영상에서 특징점들이 너무 많이 추출되는 경우 인식이나 학습단계에서의 알고리즘 수행 시간을 증가시키는 원인이 된다. 또환 중요하지 않은 특징점(배경이나 가려짐 영역, 기타 객체에서 추출된 특징점)이나 조명 변화에 민감한 영역에서 임의로(arbitrarily) 추출된 특징점은 인식률을 저하시키는 문제를 발생시킨다. 특히 도시환경에서 촬영된 영상의 특징점을 추출할 때 이러한 문제 현상들이 빈번하게 발생한다. 본 논문에서는 이러한 문제를 해결하고자 multi-view 영상에서 건물의 homography를 기반으로 정확히 정합된 특징점인 inlier만을 선택하는 알고리즘을 제안한다. Inlier로 분류된 특징점들은 건물 인식 시스템을 구성하기 위해 사용되고 조명 변화에 민감한 영역에서 임의로 추출된 특징점들은 영역 기반 특징을 추출하여 건물 인식 시스템의 인식률을 높인다. 또한 이를 이용하여 인식하고자 하는 건물과의 상관관계가 적은 잉여 영상들을 DB에서 제거하는 방법도 제안한다. 실험을 통하여 제안하는 기법의 우수성을 보였다.

• Journal of Korea Multimedia Society
• /
• v.8 no.9
• /
• pp.1195-1207
• /
• 2005

In this paper, we describe image matching techniques which is automatically retrieving the exact matching area using geometry-phase information. We proposed a Matching Method which is rapidly estimating the correspondent points between adjacent images that included big-rotation and top-bottom movement element. It is a method that reduce computation quantity to be required to find an exact correspondent position using geometry-phase information of extracted points in images and DT map which set the distance value among feature points and other points on the basis of each feature point of a image. The proposed method shows good performance especially in the part to search a exact correspondent position between adjacent images that included big-rotation and top-bottom movement element.