• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.9
• /
• pp.349-358
• /
• 2021

Recently, a method of finding the attention area or localization area for an object of an image using CAM (Class Activation Map)[1] has been variously carried out as a study of WSOL (Weakly Supervised Object Localization). The attention area extraction from the object heat map using CAM has a disadvantage in that it cannot find the entire area of the object by focusing mainly on the part where the features are most concentrated in the object. To improve this, using CAM and Selective Search[6] together, we first expand the attention area in the heat map, and a Gaussian smoothing is applied to the extended area to generate retraining data. Finally we train the data to expand the attention area of the objects. The proposed method requires retraining only once, and the search time to find an localization area is greatly reduced since the selective search is not needed in this stage. Through the experiment, the attention area was expanded from the existing CAM heat maps, and in the calculation of IOU (Intersection of Union) with the ground truth for the bounding box of the expanded attention area, about 58% was improved compared to the existing CAM.

• Journal of Advanced Navigation Technology
• /
• v.27 no.1
• /
• pp.126-132
• /
• 2023

Object recognition is a technology that detects and identifies various objects in images and videos. LBP is a descriptor that operates robustly to illumination variations and is actively used in object recognition. LBP considers the range of neighboring pixels, the order of combining the neighbors after the comparison operation, and the starting position of combining. In particular, the starting position of the LBP becomes the "most significant bit"; it dramatically affects the performance of object recognition. In this paper, based on the N starting positions, the data most similar to the input data are searched in each of the N feature spaces. Object recognition is performed by the confidence measure that can compare different results of each feature space under the same criterion and select the most reliable result. In the experimental results, it was confirmed that there is a difference in performance depending on the starting position of LBP. The proposed method showed a high performance of up to 12.66% compared to the recognition performance of the existing LBP.

• Journal of Korea Spatial Information System Society
• /
• v.7 no.3 s.15
• /
• pp.67-81
• /
• 2005

Recently, the interest and research on the development of new application services such as the Location Based Service and Telemetics providing the emergency service, neighbor information search, and route search according to the development of the Geographic Information System have been increasing. User's search in the spatio-temporal database which is used in the field of Location Based Service or Telemetics usually fixes the current time on the time axis and queries the spatial and aspatial attributes. Thus, if the range of query on the time axis is extensive, it is difficult to efficiently deal with the search operation. For solving this problem, the snapshot, a method to summarize the location data of moving objects, was introduced. However, if the range to store data is wide, more space for storing data is required. And, the snapshot is created even for unnecessary space that is not frequently used for search. Thus, non storage space and memory are generally used in the snapshot method. Therefore, in this paper, we suggests the Hash-based Spatio-Temporal Clustering Algorithm(H-STCA) that extends the two-dimensional spatial hash algorithm used for the spatial clustering in the past to the three-dimensional spatial hash algorithm for overcoming the disadvantages of the snapshot method. And, this paper also suggests the knowledge extraction algorithm to extract the knowledge for the path search of moving objects from the past location data based on the suggested H-STCA algorithm. Moreover, as the results of the performance evaluation, the snapshot clustering method using H-STCA, in the search time, storage structure construction time, optimal path search time, related to the huge amount of moving object data demonstrated the higher performance than the spatio-temporal index methods and the original snapshot method. Especially, for the snapshot clustering method using H-STCA, the more the number of moving objects was increased, the more the performance was improved, as compared to the existing spatio-temporal index methods and the original snapshot method.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.3
• /
• pp.28-36
• /
• 2009

In this paper, we propose a frame-rate up-conversion method for temporal quality enhancement. The proposed method adaptively changes search range during hierarchical motion estimation and reconstructs hole regions using the proposed bi-direction prediction and linear interpolation. In order to alleviate errors due to inaccurate motion vector estimation, search range is adaptively changed based on reliability and for more accurate, motion estimation is performed in descending order of block variance. After segmentation of background and object regions, for filling hole regions, the pixel values of background regions are reconstructed using linear interpolation and those of object regions are compensated based on the proposed hi-directional prediction. The proposed algorithm is evaluated in terms of PSNR with original uncompressed sequences. Experimental results show that the proposed algorithm is better than conventional methods by around 2dB, and blocky artifacts and blur artifacts are significantly diminished.

• Journal of Korean Society for Geospatial Information Science
• /
• v.15 no.3
• /
• pp.27-32
• /
• 2007

3D based information is demanded increasingly as cities grow three dimensionally and buildings become large and complex. The use of 3D GIS is also getting attention as fundamental data for ubiquitous computing applications such as location-based guidance, path finding and emergency escaping. However, most 3D modeling techniques are focused on the visualization of buildings or terrains and do not have topological structures required in spatial analyses. In this paper, we introduce a method to incorporate topological relationship into 3D models by combining 2D GIS layers and 3D model. We divide indoor spaces of a 3D model into discrete objects and then define the relationship with corresponding features in 2D GIS layers through database records. We also show how to construct hallways network in the 2D-3D integrated building model. Finally, we test different cases of route finding situations inside a building such as normal origin-destination path finding and emergency evacuation.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2015.10a
• /
• pp.695-698
• /
• 2015

In this paper proposes bat tracking in baseball broadcasting using CAMshift and Kalman filter. The bat is changing fast during the swing, the shape also continues to rotate. For this reason, to apply the CAMshift to self adjust the size of the search window in order to use the color information to the invariant of the bat. Because it uses the color information if there are objects of similar color to the background because of the interruption on the track narrows the search range in range of motion detection by using the MHI(Motion History Image). By applying a Kalman filter, limit changing on the size of the search window, and it can be obtained higher track accuracy. But, this proposed method was limited color change by light.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.30 no.2
• /
• pp.125-134
• /
• 2012

In this paper, we propose a method to analyze M:N corresponding relations in semantic matching, especially focusing on feature class matching. Similarities between any class pairs are measured by spatial objects which coexist in the class pairs, and corresponding classes are obtained by clustering with these pairwise similarities. We applied a graph embedding method, which constructs a global configuration of each class in a low-dimensional Euclidean space while preserving the above pairwise similarities, so that the distances between the embedded classes are proportional to the overall degree of similarity on the edge paths in the graph. Thus, the clustering problem could be solved by employing a general clustering algorithm with the embedded coordinates. We applied the proposed method to polygon object layers in a topographic map and land parcel categories in a cadastral map of Suwon area and evaluated the results. F-measures of the detected class pairs were analyzed to validate the results. And some class pairs which would not detected by analysis on nominal class names were detected by the proposed method.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.15 no.11
• /
• pp.2321-2326
• /
• 2011

Cervical vertebrae are a complex structure and an important part of human body connecting the head and the trunk. In this paper, we propose a method to extract sternocleidomastoid muscle from ultrasonography images of cervical vertabrae automatically. In our method, Region of Interests(ROI) is extracted first from an ultrasonography image after removing unnecessary auxiliary information such as metrics. Then we apply Ends-in search stretching algorithm in order to enhance the contrast of brightness. Average binarization is then applied to those pixels which its brightness is sufficiently large. The noise part is removed by image processing algorithms. After extracting fascia encloses sternocleidomastoid muscle, target muscle object is extracted using the location information of fascia according to the number of objects in the fascia. When only one object is to be extracted, we search downward first to extract the target muscle area and then search from right to left to extract the area and merge them. If there are two target objects, we extract first from the upper-bound of higher object to the lower-bound of lower object and then remove the fascia of the target object area. Smearing technique is used to restore possible loss of the fat area in the process. The thickness of sternocleidomastoid muscle is then calculated as the maximum thickness of those extracted objects. In this experiment with 30 real world ultrasonography images, the proposed method verified its efficacy and accuracy by health professionals.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.9
• /
• pp.2173-2184
• /
• 1997

Among spatial queries handled in spatial database systems, nearest neighbor queries to find the nearest spatial object from the given locaion occur frequently. The number of searched nodes in an index must be minimized in order to increase the performance of nearest neighbor queries. An Existing approach considered only the processing of an nearest neighbor query in a two-dimensional search space and could not optimize the number of searched nodes accurately. In this paper, we propose the optimal search distance and prove its properties. The proposed optimal search distance is the measurement of a new search distance for accurately selecting the nodes which will be searched in processing nearest neighbor queries. We present an algorithm for processing the nearest neighbor query by applying the optimal search distance to R-trees and prove that the result of query processing is correcter than the existing approach.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 1996.06a
• /
• pp.169-174
• /
• 1996

객체 인식은 대용량의 영상 데이터를 분석, 탐색하고 재구성하기 위한 내용기반 영상 검색의 매우 중요한 분야이며, 특히 인간의 얼굴은 검색 영상 내에서 대부분 주요한 장면에 위치하고 있기 때문에 그 비중이 매우 크다. 본 논문에서는 내용기반 영상 검색을 위한 실시간 얼굴 영역 추출 방법을 제안한다. 제안된 방법에서는 다층 피라미드 구조와 간단한 형태의 머리 형판을 사용하여 얼굴의 후보 영역을 추출한 후, 보다 정확한 얼굴 영역을 추출하기 위하여 후보 영역 내에서 눈의 위치를 탐색하고, 두 눈의 위치를 기준으로 최종적인 얼굴 영역을 추출하였다. 얼굴 후보 영역 추출 단계에서는 얼굴의 형태 정보를 포함하고 있는 모자이크 형판을 사용하여 머리와 턱을 포함한 얼굴 영역을 추출하였으며, 눈 위치 추출 단계에서는 눈의 위치 정보를 사용하여 눈의 탐색 영역을 결정하고, 탐색 영역 내에서 이진 영상 형판을 사용하여 눈의 위치를 추출한 후, 눈 영역의 무게 중심을 눈의 중심 위치로 설정하였다. 마지막 얼굴 영역 추출단계에서는 두 눈의 위치를 기준으로 사각형의 영역을 얼굴 영역으로 추출하였다. 제안된 방법의 성능을 검증하기 위하여 1700장의 다양한 영상에 대하여 실험하였으며, 실험 결과 한 장의 영상에서 얼굴 영역을 추출하는데 있어서, Pentium 166Mz의 PC상에서 평균 3.2초의 처리 속도와 91.7%의 추출률을 보임으로써, 실시간 얼굴 영역 추출에 매우 효과적임을 알 수 있었다.