• Journal of Korea Multimedia Society
• /
• v.8 no.10
• /
• pp.1306-1313
• /
• 2005

Relevance Feedback(RF), a method to use perceptual feedback in image retrieval, refines a query by the relevance information from a user. However, the user's feedback information is thrown away as soon as a search session ends. So, this paper proposes an enhanced version of RF, which is designed to accumulate human perceptual responses over time through relevance feedback and to dynamically combine the accumulated high-level relevance information with low-level features to further improve the retrieval effectiveness. Experimental results are presented to prove the potential of the proposed RF.

• The KIPS Transactions:PartB
• /
• v.11B no.6
• /
• pp.701-708
• /
• 2004

Existing contents-based video retrieval systems search by using a single method such as annotation-based or feature-based retrieval. Hence, it not only shows low search efficiency, but also requires many efforts to provide system administrator or annotator with a perfect automatic processing. Tn this paper, we propose an agent-based, and automatic and unified semantics-based video retrieval system, which support various semantics-retrieval of the massive video data by integrating the feature-based retrieval and the annotation-based retrieval. The indexing agent embodies the semantics about annotation of extracted key frames by analyzing a fundamental query of a user and by selecting a key-frame image that is ed by a query. Also, a key frame selected by user takes a query image of the feature-based retrieval and the indexing agent searches and displays the most similar key-frame images after comparing query images with key frames in the database by using the color-multiple-partition histogram techniques. Furthermore, it is shown that the performance of the proposed system can be significantly improved.

• Journal of Korea Multimedia Society
• /
• v.8 no.2
• /
• pp.269-278
• /
• 2005

In order to process video data effectively, it is required that the content information of video data is loaded in database and semantics-based retrieval method can be available for various queries of users. In this paper, we propose a video retrieval system which support semantics retrieval of various users for massive video data by user's keywords and comparison area learning based on automatic agent. By user's fundamental query and selection of image for key frame that extracted from query, the agent gives the detail shape for annotation of extracted key frame. Also, key frame selected by user becomes a query image and searches the most similar key frame through color histogram comparison and comparison area learning method that proposed. From experiment, the designed and implemented system showed high precision ratio in performance assessment more than 93 percents.

• Journal of The Institute of Information and Telecommunication Facilities Engineering
• /
• v.9 no.2
• /
• pp.73-79
• /
• 2010

The most popular technique for image retrieval in a heterogeneous collection of color images is the comparison of images based on their color histogram. The color histogram describes the distribution of colors in the color space of a color image. In the most image retrieval systems, the color histogram is used to compute similarities between the query image and all the images in a database. But, small changes in the resolution, scaling, and illumination may cause important modifications of the color histogram, and so two color images may be considered to be very different from each other even though they have completely related semantics. A new method of color feature representation based on the 3-dimensional RGB color map is proposed to improve the defects of the color histogram. The proposed method is based on the three 2-dimensional projection map evaluated by projecting the RGB color space on the RG, GB, and BR surfaces. The experimental results reveal that the proposed is less sensitive to small changes in the scene and that achieve higher retrieval performances than the traditional color histogram.

• Journal of information and communication convergence engineering
• /
• v.2 no.2
• /
• pp.123-127
• /
• 2004

The most popular technique for image retrieval in a heterogeneous collection of color images is the comparison of images based on their color histogram. The color histogram describes the distribution of colors in the color space of a color image. In the most image retrieval systems, the color histogram is used to compute similarities between the query image and all the images in a database. But, small changes in the resolution, scaling, and illumination may cause important modifications of the color histogram, and so two color images may be considered to be very different from each other even though they have completely related semantics. A new method of color feature representation based on the 3-dimensional RGB color map is proposed to improve the defects of the color histogram. The proposed method is based on the three 2-dimensional projection map evaluated by projecting the RGB color space on the RG, GB, and BR surfaces. The experimental results reveal that the proposed is less sensitive to small changes in the scene and that achieve higher retrieval performances than the traditional color histogram.

• ETRI Journal
• /
• v.27 no.3
• /
• pp.326-329
• /
• 2005

The MPEG-4 multiple auxiliary component (MAC) is a good mechanism to achieve one-stream stereoscopic video coding. However, there is no syntax or semantics for the residual texture data of the disparity-compensated image in the current MAC. Therefore, we propose a novel disparity-compensated coding method using the MAC for stereoscopic video. We also define a novel MAC semantics in MPEG-4 so as to support the proposed coding algorithm. The major difference between the existing and proposed coding methods using the MAC is the addition of the residual texture coding.

• Annual Conference of KIPS
• /
• 2004.05a
• /
• pp.35-38
• /
• 2004

기존의 내용기반 이미지 검색 시스템은 색상, 질감, 모양등과 같은 특징 벡터를 추출하여 검색하는 방법이 많이 연구되어 왔다. 특히 색상 정보는 이미지를 검색하기 위하여 중요한 정보로 사용되고 있다. 따라서 색상 이미지를 검색하기 위해서 평균 RGB, HSI값을 이용하거나 히스토그램을 이용하는 방식이 많이 사용 되어왔다. 본 논문에서는 사람이 시각적으로 보고 느끼는 색상(H), 채도(S), 명도(I) 방식을 이용한 HSI값을 사용하여 색상 의미를 이용한 지능적 이미지 검색 기법을 제안하고 알고리즘을 설명한다. 색상 의미(Color Semantics)란 사람의 시각적인 특징을 기반으로 칼라 이미지에 적용하여 감성 형용사 기반으로 검색할 수 있는 방법이다. 색상 의미를 이용한 지능적 이미지 검색은 색상-기반 질의(color-based retrieval)를 제공할 뿐만 아니라 인간의 감성이나 느낌에 의한 의미-기반 질의(semantic-based retrieval)방식을 가능하게 한다. 즉, "시원한 이미지" 혹은 "부드러운 이미지"를 검색하는 방식이다. 따라서 사용자의 검색 의도를 보다 정확하게 표현할 수 있으며, 검색의 결과에 대한 만족도를 향상 시킬 수 있다.

• Annual Conference of KIPS
• /
• 2019.05a
• /
• pp.401-403
• /
• 2019

In this paper, we propose a generative adversarial networks (GAN) based text-to-image generating method. In many natural language processing tasks, which word expressions are determined by their term frequency -inverse document frequency scores. Word2Vec is a type of neural network model that, in the case of an unlabeled corpus, produces a vector that expresses semantics for words in the corpus and an image is generated by GAN training according to the obtained vector. Thanks to the understanding of the word we can generate higher and more realistic images. Our GAN structure is based on deep convolution neural networks and pixel recurrent neural networks. Comparing the generated image with the real image, we get about 88% similarity on the Oxford-102 flowers dataset.

• ETRI Journal
• /
• v.29 no.5
• /
• pp.700-702
• /
• 2007

We present a relevance feedback approach based on multi-class support vector machine (SVM) learning and cluster-merging which can significantly improve the retrieval performance in region-based image retrieval. Semantically relevant images may exhibit various visual characteristics and may be scattered in several classes in the feature space due to the semantic gap between low-level features and high-level semantics in the user's mind. To find the semantic classes through relevance feedback, the proposed method reduces the burden of completely re-clustering the classes at iterations and classifies multiple classes. Experimental results show that the proposed method is more effective and efficient than the two-class SVM and multi-class relevance feedback methods.

• Journal of Multimedia Information System
• /
• v.1 no.1
• /
• pp.55-67
• /
• 2014

Image topic and emotion analysis is an important component of online image retrieval, which nowadays has become very popular in the widely growing social media community. However, due to the gaps between images and texts, there is very limited work in literature to detect one image's Topics and Emotions in a unified framework, although topics and emotions are two levels of semantics that often work together to comprehensively describe one image. In this work, a unified model, Joint Topic/Emotion Multi-Modal Hierarchical Latent Dirichlet Allocation (JTE-MMHLDA) model, which extends previous LDA, mmLDA, and JST model to capture topic and emotion information at the same time from heterogeneous data, is proposed. Specifically, a two level graphical structured model is built to realize sharing topics and emotions among the whole document collection. The experimental results on a Flickr dataset indicate that the proposed model efficiently discovers images' topics and emotions, and significantly outperform the text-only system by 4.4%, vision-only system by 18.1% in topic detection, and outperforms the text-only system by 7.1%, vision-only system by 39.7% in emotion detection.