• Proceedings of the KIEE Conference
• /
• 2001.07d
• /
• pp.2656-2659
• /
• 2001

본 논문은 교사 있는 학습기의 Kohonen Feature Map에 고차 뉴런을 도입, 고차 뉴런을 이용한 Kohonen의 자기 조직화 맵을 제안한다. 일반적인 Kohonen Feature Map의 특징은 입력신호를 받아 출력 면(Kohonen Feature Map) 내의 특정한 위치 주위에 집중하는 메커니즘으로 즉, 국소집중 반응을 구하는 구조이다. 본 논문에서는 종래형의 Kohonen Feature Map의 특징을 보유하며 교사 있는 학습기의 Kohonen Feature Map에 고차 뉴런을 도입하여 국소집중반응 및 특징 축출이 용이하도록 네트워크 구조를 개선한 것이다. 특히, 일차 뉴런의 문제점인 비선형 분리 문제에 대하여 교사 있는 학습기의 Kohonen Feature Map의 입력층에 고차 뉴런을 도입함으로 비선형 분리 가능한 형태의 네트워크 구조로 형성하였다. 그러나, 일반적인 고차 뉴런의 문제점을 보안하기 위해 본 논문에서는 오직 2차 뉴런만을 생성하였으며 중복되는 뉴런을 최대한 억제하였다. 본 제안 모델의 특성을 살펴보기 위해 XOR문제와 20개의 Alphabet을 식별하는 패턴인식 시뮬레이션을 했으며, 본 제안 모델의 범화능력을 알아보기 위하여 Mirror Symmetry를 사용하여 계산기 시뮬레이션을 했다. 그 결과, 본 제안 모델이 종래형의 네트워크 구조보다 뛰어난 인식률을 얻을 수 있었다.

• Journal of applied mathematics & informatics
• /
• v.9 no.2
• /
• pp.793-806
• /
• 2002

Feature-based similarity retrieval become an important research issue in image database systems. The features of image data are useful to discrimination of images. In this paper, we propose the highspeed k-Nearest Neighbor search algorithm based on Self-Organizing Maps. Self-Organizing Maps (SOM) provides a mapping from high dimensional feature vectors onto a two-dimensional space. The mapping preserves the topology of the feature vectors. The map is called topological feature map. A topological feature map preserves the mutual relations (similarity) in feature spaces of input data. and clusters mutually similar feature vectors in a neighboring nodes. Each node of the topological feature map holds a node vector and similar images that is closest to each node vector. In topological feature map, there are empty nodes in which no image is classified. We experiment on the performance of our algorithm using color feature vectors extracted from images. Promising results have been obtained in experiments.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.283-294
• /
• 2022

The feature map used in the network for deep learning generally has larger data than the image and a higher compression rate than the image compression rate is required to transmit the feature map. This paper proposes a method for transmitting a pyramid feature map with high compression rate, which is used in a network with an FPN structure that has robustness to object size in deep learning-based image processing. In order to efficiently compress the pyramid feature map, this paper proposes a structure that predicts a pyramid feature map of a level that is not transmitted with pyramid feature map of some levels that transmitted through the proposed prediction network to efficiently compress the pyramid feature map and restores compression damage through the proposed reconstruction network. Suggested mAP, the performance of object detection for the COCO data set 2017 Train images of the proposed method, showed a performance improvement of 31.25% in BD-rate compared to the result of compressing the feature map through VTM12.0 in the rate-precision graph, and compared to the method of performing compression through PCA and DeepCABAC, the BD-rate improved by 57.79%.

• Proceedings of the KIEE Conference
• /
• 2004.05a
• /
• pp.141-143
• /
• 2004

This paper presents an algorithm in which a feature map is built and localization of a mobile robot is carried out for indoor environments. The algorithm proposes an approach which extracts scale-invariant features of natural landmarks from a pair of stereo images. The feature map is built using these features and updated by merging new landmarks into the map and removing transient landmarks over time. And the position of the robot in the map is estimated by comparing with the map in a database by means of an Extended Kalman filter. This algorithm is implemented and tested using a Pioneer 2-DXE and preliminary results are presented in this paper.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.52 no.5
• /
• pp.277-282
• /
• 2003

In this paper we propose Using Higher Order Neuron on the Supervised Learning Machine of the Kohonen Feature Map. The architecture of proposed model adopts the higher order neuron in the input layer of Kohonen Feature Map as a Supervised Learning Machine. It is able to estimate boundary on input pattern space because or the higher order neuron. However, it suffers from a problem that the number of neuron weight increases because of the higher order neuron in the input layer. In this time, we solved this problem by placing the second order neuron among the higher order neuron. The feature of the higher order neuron can be mapped similar inputs on the Kohonen Feature Map. It also is the network with topological mapping. We have simulated the proposed model in respect of the recognition rate by XOR problem, discrimination of 20 alphabet patterns, Mirror Symmetry problem, and numerical letters Pattern Problem.

• Proceedings of the KSRS Conference
• /
• 2008.10a
• /
• pp.433-436
• /
• 2008

A demand on the spatial data management has been rapidly increased with the introduction and diffusion process of ITS, Telematics, and Wireless Sensor Network, and many different people use the digital map that offers various thematic spatial data. Spatial data for digital map can manage to tile-based and feature-based data. The existing tile-based digital map management systems have difficult problems of data construction, history management, and updating based on a spatial object. In order to solve these problems, this paper proposed the data model for the feature-based digital map management system that is designed for feature-based seamless map, history management, real-time updating of spatial data, and analyzed the validity and utility of the proposed model.

• The Journal of Korea Robotics Society
• /
• v.9 no.2
• /
• pp.96-103
• /
• 2014

Global localization is one of the essential issues for mobile robot navigation. In this study, an indoor global localization method is proposed which uses a Kinect sensor and a monocular upward-looking camera. The proposed method generates an environment map which consists of a grid map, a ceiling feature map from the upward-looking camera, and a spatial feature map obtained from the Kinect sensor. The method selects robot pose candidates using the spatial feature map and updates sample poses by particle filter based on the grid map. Localization success is determined by calculating the matching error from the ceiling feature map. In various experiments, the proposed method achieved a position accuracy of 0.12m and a position update speed of 10.4s, which is robust enough for real-world applications.

• Journal of Digital Convergence
• /
• v.10 no.3
• /
• pp.173-178
• /
• 2012

In this paper, a new saliency map generation method is discussed which extracts objects effectively using extracted Salient Region. Feature map is constructed first using four features of edge, hue of HSV color model, focus and entropy and then conspicuity map is generated from Center Surround Differences using the feature map. Final saliency map is constructed by the combination of conspicuity maps. Saliency map generated using this procedure is compared to the conventional technique and confirmed that new technique has better results.

• Proceedings of the KAIS Fall Conference
• /
• 2003.11a
• /
• pp.223-230
• /
• 2003

Feature-based similarity retrieval has become an important research issue in multimedia database systems. The features of multimedia data are useful for discriminating between multimedia objects (e 'g', documents, images, video, music score, etc.). For example, images are represented by their color histograms, texture vectors, and shape descriptors, and are usually high-dimensional data. The performance of conventional multidimensional data structures(e'g', R- Tree family, K-D-B tree, grid file, TV-tree) tends to deteriorate as the number of dimensions of feature vectors increases. The R＊-tree is the most successful variant of the R-tree. In this paper, we propose a SOM-based R＊-tree as a new indexing method for high-dimensional feature vectors.The SOM-based R＊-tree combines SOM and R＊-tree to achieve search performance more scalable to high dimensionalities. Self-Organizing Maps (SOMs) provide mapping from high-dimensional feature vectors onto a two dimensional space. The mapping preserves the topology of the feature vectors. The map is called a topological of the feature map, and preserves the mutual relationship (similarity) in the feature spaces of input data, clustering mutually similar feature vectors in neighboring nodes. Each node of the topological feature map holds a codebook vector. A best-matching-image-list. (BMIL) holds similar images that are closest to each codebook vector. In a topological feature map, there are empty nodes in which no image is classified. When we build an R＊-tree, we use codebook vectors of topological feature map which eliminates the empty nodes that cause unnecessary disk access and degrade retrieval performance. We experimentally compare the retrieval time cost of a SOM-based R＊-tree with that of an SOM and an R＊-tree using color feature vectors extracted from 40, 000 images. The result show that the SOM-based R＊-tree outperforms both the SOM and R＊-tree due to the reduction of the number of nodes required to build R＊-tree and retrieval time cost.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.43 no.5 s.311
• /
• pp.12-21
• /
• 2006

In this paper, we propose a technique for extracting attentive objects in images using feature maps, regardless of the complexity of images and the position of objects. The proposed method uses feature maps with edge and color information in order to extract attentive objects. We also propose a reference map which is created by integrating feature maps. In order to create a reference map, feature maps which represent visually attentive regions in images are constructed. Three feature maps including edge map, CbCr map and H map are utilized. These maps contain the information about boundary regions by the difference of intensity or colors. Then the combination map which represents the meaningful boundary is created by integrating the reference map and feature maps. Since the combination map simply represents the boundary of objects we extract the candidate object regions including meaningful boundaries from the combination map. In order to extract candidate object regions, we use the convex hull algorithm. By applying a segmentation algorithm to the area of candidate regions to separate object regions and background regions, real object regions are extracted from the candidate object regions. Experiment results show that the proposed method extracts the attentive regions and attentive objects efficiently, with 84.3% Precision rate and 81.3% recall rate.