• Proceedings of the Korea Information Processing Society Conference
• /
• 2020.05a
• /
• pp.555-558
• /
• 2020

본 논문에서는 입력 영상에 담긴 다양한 물체들과 그들 간의 관계를 효과적으로 탐지하여, 하나의 장면 그래프로 표현해내는 새로운 심층 신경망 모델을 제안한다. 제안 모델에서는 물체와 관계의 효과적인 탐지를 위해, 합성 곱 신경망 기반의 시각 맥락 특징들뿐만 아니라 언어 맥락 특징들을 포함하는 다양한 멀티 모달 맥락 정보들을 활용한다. 또한, 제안 모델에서는 관계를 맺는 두 물체 간의 상호 의존성이 그래프 노드 특징값들에 충분히 반영되도록, 그래프 신경망을 이용해 맥락 정보를 임베딩한다. 본 논문에서는 Visual Genome 벤치마크 데이터 집합을 이용한 비교 실험들을 통해, 제안 모델의 효과와 성능을 입증한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.11a
• /
• pp.646-647
• /
• 2022

Multiview stereo (MVS) 3D reconstruction of a scene from images is a fundamental computer vision problem that has been thoroughly researched in recent times. Traditionally, MVS approaches create dense correspondences by constructing regularizations and hand-crafted similarity metrics. Although these techniques have achieved excellent results in the best Lambertian conditions, traditional MVS algorithms still contain a lot of artifacts. Therefore, in this study, we suggest using a transformer network to accelerate the MVS reconstruction. The network is based on a transformer model and can extract dense features with 3D consistency and global context, which are necessary to provide accurate matching for MVS.

• Journal of Information Processing Systems
• /
• v.14 no.2
• /
• pp.469-486
• /
• 2018

We propose an enhanced version of the local binary pattern (LBP) operator for texture extraction in images in the context of image retrieval. The novelty of our proposal is based on the observation that the LBP exploits only the lowest kind of local information through the global histogram. However, such global Histograms reflect only the statistical distribution of the various LBP codes in the image. The block based LBP, which uses local histograms of the LBP, was one of few tentative to catch higher level textural information. We believe that important local and useful information in between the two levels is just ignored by the two schemas. The newly developed method: gradual locality integration of binary patterns (GLIBP) is a novel attempt to catch as much local information as possible, in a gradual fashion. Indeed, GLIBP aggregates the texture features present in grayscale images extracted by LBP through a complex structure. The used framework is comprised of a multitude of ellipse-shaped regions that are arranged in circular-concentric forms of increasing size. The framework of ellipses is in fact derived from a simple parameterized generator. In addition, the elliptic forms allow targeting texture directionality, which is a very useful property in texture characterization. In addition, the general framework of ellipses allows for taking into account the spatial information (specifically rotation). The effectiveness of GLIBP was investigated on the Corel-1K (Wang) dataset. It was also compared to published works including the very effective DLEP. Results show significant higher or comparable performance of GLIBP with regard to the other methods, which qualifies it as a good tool for scene images retrieval.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2013.10a
• /
• pp.1015-1017
• /
• 2013

Object tracking, in general, is a challenging problem. Difficulties in tracking objects can arise due to abrupt object motion, changing appearance patterns of both the object and the scene, non-rigid object structures, object-to-object and object-to-scene occlusions, and camera motion. Tracking is usually performed in the context of higher-level applications that require the location or the shape of the object in every frame. This paper describes an object tracking system based on active vision with two cameras, into algorithm of single camera tracking system an object active visual tracking and object locked system based on Extend Kalman Filter (EKF) is introduced, by analyzing data from which the next running state of the object can be figured out and after the tracking is performed at each of the cameras, the individual tracks are to be fused (combined) to obtain the final system object track.

• Journal of Advanced Navigation Technology
• /
• v.17 no.6
• /
• pp.792-797
• /
• 2013

4D film is just a film that made by adding some physical effects to 3D film or general film. In order to provide physical effects to the audience, the data that make the physical effect must be added to each frames. In this paper, we proposed a video detection system that can efficiently provide physical effects by assessing the present situation such as explosion scene, snowing scene. The proposed video detection system contains an algorithm for fire detection by using R color and $C_r$ value, and also an algorithm for snow detection by using RGB color model. The system constitutes in a MCU that from 8051 family. In the performance evaluations, the result shows that 91% of detection rate in case of fire and 25% of false detection rate in case of snow. Also the system is capable of providing physical effects automatically.

• Science of Emotion and Sensibility
• /
• v.13 no.1
• /
• pp.251-258
• /
• 2010

'Naturalness' is the important "ness" which is a key factor in image quality assessment. 'Naturalness' is a representive factor depending on the context of the image which arouses different emotions. The Image Quality Circle was split into two steps. The first step is predicting the visual perceptual attribute which are lightness, colourfulness, hue and contrast. The next step is SSE which is dependent to image contents. In this study the image contents are grouped in genres. The images were rendered using four different colour attributes which are lightness, contrast, colourfulness and hue. Using a scale, the score of image quality and SSE was asked to each participant for all rendered images. A seven-point category scale of increasing amount of "ness" is used as a quantitative adjectives sequence. The image quality model was built by combining the SSEs for each scene. The SSEs, where vividness is common, are considered as independent variables to predict the image quality score. Then the vividness model was built using colour attributes as variables to predict the vividness of each scene (genre). Vividness is an important factor of naturalness which the meaning is different for all scenes that links the naturalness and image quality. The vividness meaning was different for each scene (genre). Therefore, the colour attributes that express the vividness would depend on the image content.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.4
• /
• pp.737-743
• /
• 2017

The intelligent life-log system proposed in this paper is intended to identify and record a myriad of everyday life information as to the occurrence of various events based on when, where, with whom, what and how, that is, a wide variety of contextual information involving person, scene, ages, emotion, relation, state, location, moving route, etc. with a unique tag on each piece of such information and to allow users to get a quick and easy access to such information. Context awareness generates and classifies information on a tag unit basis using the auto-tagging technology and biometrics recognition technology and builds a situation information database. In this paper, we developed an active modeling method and an application that recognizes expressionless and smile expressions using lip lines to automatically record emotion information.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.35 no.5C
• /
• pp.464-471
• /
• 2010

The demand for high quality of multimedia applications, which far surpasses the rapid evolution of transmission and storage technologies, makes better compression coding capabilities ever increasingly more important. In order to provide enhanced video coding performance, this paper proposes an efficient competition-based motion vector coding scheme. The proposed algorithm adaptively forms the motion vector predictors based on the contexts of scene characteristics such as camera motion and nearby motion vectors, providing more efficient candidate predictors than the previous competition-based motion vector coding schemes which resort to the fixed candidates optimized by extensive simulations. Up to 200% of compression gain was observed in the experimental results for the proposed scheme applied to the motion vector selection for skip mode processing.

• KIPS Transactions on Software and Data Engineering
• /
• v.7 no.2
• /
• pp.69-76
• /
• 2018

Referring expressions are natural language constructions used to identify particular objects within a scene. In this paper, we propose a new deep neural network model for referring expression comprehension. The proposed model finds out the region of the referred object in the given image by making use of the rich information about the referred object itself, the context object, and the relationship with the context object mentioned in the referring expression. In the proposed model, the object matching score and the relationship matching score are combined to compute the fitness score of each candidate region according to the structure of the referring expression sentence. Therefore, the proposed model consists of four different sub-networks: Language Representation Network(LRN), Object Matching Network (OMN), Relationship Matching Network(RMN), and Weighted Composition Network(WCN). We demonstrate that our model achieves state-of-the-art results for comprehension on three referring expression datasets.

• Journal of the Korean Institute of Traditional Landscape Architecture
• /
• v.29 no.4
• /
• pp.49-58
• /
• 2011

This article attempts to propose the control planning of townscape color around the historic and cultural heritages. The streets and roads of historic conservation and the changing perspectives to which it gives rise in Stockholm has changed dramatically during this century. New development or changes to existing buildings should be carried out in a way which acknowledges its surroundings and is a good neighbour, both in the cultural and social sense that makes good color design sense. There are many examples of townscape color, but the conservative and the historical streets and roads in the whole of the Stockholm city should benefit from careful design of the environment. To achieve this purposes, some strategies of case study of several streets and roads are reviewed; designing color context to relate to urban architectural design proposals of specific sites of cultural heritages are explored. In all new developments the scale of new buildings and the material finishes and colors used should respect the character of their surroundings and have due regard to the setting of any listed building. Streetscape color of visual assessment proposals should aim to help assimilate the development into the local scene. Important streets and roads should also include color townscape.