• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제9권4호
• /
• pp.1424-1440
• /
• 2015

It is a challenging problem to search the intended images from a large number of candidates. Content based image retrieval (CBIR) is the most promising way to tackle this problem, where the most important topic is to measure the similarity of images so as to cover the variance of shape, color, pose, illumination etc. While previous works made significant progresses, their adaption ability to dataset is not fully explored. In this paper, we propose a similarity learning method on the basis of probabilistic generative model, i.e., probabilistic latent semantic analysis (PLSA). It first derives Fisher kernel, a function over the parameters and variables, based on PLSA. Then, the parameters are determined through simultaneously maximizing the log likelihood function of PLSA and the retrieval performance over the training dataset. The main advantages of this work are twofold: (1) deriving similarity measure based on PLSA which fully exploits the data distribution and Bayes inference; (2) learning model parameters by maximizing the fitting of model to data and the retrieval performance simultaneously. The proposed method (PLSA-FK) is empirically evaluated over three datasets, and the results exhibit promising performance.

• 한국정보통신학회논문지
• /
• 제6권8호
• /
• pp.1358-1364
• /
• 2002

현재 방송 및 인터넷분야에서는 멀티미디어 정보가 급격히 증가하고 있다. 본 논문에서는 멀티미디어 정보 중에서 정지영상 검색을 위해 사용자가 질의(query)를 원하는 물체영역을 선택한 후 유사물체를 영상 데이터베이스 내에서 검색할 수 있는 내용기반 영상검색 시스템을 구현하였다. 질의영상으로부터 우선 컬러특성을 추출하기 위해 제안한 방법으로 색상을 HSV 변환한 후 히스토그램을 구해 데이터베이스영상과 히스토그램 인터섹션을 통해 유사치를 구한다 또한 질의영상을 그레이영상으로도 변환시켜 웨블릿 변환한 후 밴디드 오토코릴로그램과 GLCM을 통해 공간적 그레이분포와 질감특성을 추출하여 유사치를 구한다. 그리고 2개의 유사치를 더하여 최종 유사도를 결정하는데 이때 각 유사치에 가중치를 적용하였다. 질의영상으로부터 컬러영상 특성뿐만 아니라 그레이영상 특성도 파악하여 단점을 보완하였고 실험결과에서도 소환성(recall) 및 정확성(precision)이 향상됨을 볼 수 있었다. 또한 가중치를 적용함으로써 검색효율이 개선되었다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제8권8호
• /
• pp.2895-2912
• /
• 2014

Content-based image retrieval has been the most important technique for managing huge amount of images. The fundamental yet highly challenging problem in this field is how to measure the content-level similarity based on the low-level image features. The primary difficulties lie in the great variance within images, e.g. background, illumination, viewpoint and pose. Intuitively, an ideal similarity measure should be able to adapt the data distribution, discover and highlight the content-level information, and be robust to those variances. Motivated by these observations, we in this paper propose a probabilistic similarity learning approach. We first model the distribution of low-level image features and derive the free energy kernel (FEK), i.e., similarity measure, based on the distribution. Then, we propose a learning approach for the derived kernel, under the criterion that the kernel outputs high similarity for those images sharing the same class labels and output low similarity for those without the same label. The advantages of the proposed approach, in comparison with previous approaches, are threefold. (1) With the ability inherited from probabilistic models, the similarity measure can well adapt to data distribution. (2) Benefitting from the content-level hidden variables within the probabilistic models, the similarity measure is able to capture content-level cues. (3) It fully exploits class label in the supervised learning procedure. The proposed approach is extensively evaluated on two well-known databases. It achieves highly competitive performance on most experiments, which validates its advantages.

• 반도체디스플레이기술학회지
• /
• 제7권4호
• /
• pp.13-18
• /
• 2008

This paper explores a content-based image retrieval framework with relevance feedback based on genetic algorithm (GA). This framework adopts GA to learn the user preferences using the similarity functions defined for all available descriptors. The objective of the GA-based learning methods is to learn the user preferences using the similarity functions and to find a descriptor combination function that best represents the user perception. Experiments were performed to validate the proposed frameworks. The experiments employed the natural image databases and color and texture descriptors to represent the content of database images. The proposed frameworks were compared with the other two relevance feedback methods regarding effectiveness in image retrieval tasks. Experiment results demonstrate the superiority of the proposed method.

• Journal of Electrical Engineering and Technology
• /
• 제8권5호
• /
• pp.1188-1193
• /
• 2013

In content based image retrieval (CBIR) system, the images are represented based upon its feature such as color, texture, shape, and spatial relationship etc. In this paper, we propose a MRI Image Retrieval using wavelet transform with mahalanobis distance measurement. Wavelet transformation can also be easily extended to 2-D (image) or 3-D (volume) data by successively applying 1-D transformation on different dimensions. The proposed algorithm has tested using wavelet transform and performance analysis have done with HH and $H^*$ elimination methods. The retrieval image is the relevance between a query image and any database image, the relevance similarity is ranked according to the closest similar measures computed by the mahalanobis distance measurement. An adaptive similarity synthesis approach based on a linear combination of individual feature level similarities are analyzed and presented in this paper. The feature weights are calculated by considering both the precision and recall rate of the top retrieved relevant images as predicted by our enhanced technique. Hence, to produce effective results the weights are dynamically updated for robust searching process. The experimental results show that the proposed algorithm is easily identifies target object and reduces the influence of background in the image and thus improves the performance of MRI image retrieval.

• 전자공학회논문지CI
• /
• 제39권4호
• /
• pp.28-37
• /
• 2002

본 논문에서는, 확률적 방법과 질의 위치 이동 방법을 융합하여 검색 성능을 향상시키는 영상검색 방법을 제안한다. 제안한 알고리즘은, 질의 영상과 데이터베이스 영상 사이의 유사도 계산에서, 확률적 방법의 유사도와 질의 위치 이동 방법의 유사도를 융합한다. 본 논문에서 이용된 확률적 방법은 부정적 예제들을 다루기에 적합하다. 반면에, 질의 위치 이동 방법은 긍정적예제의 통계적인 특성을 다룬다. 본 논문의 목적은 이러한 두 방법을 융합함으로써, 각각의 방법이 가지고 있는 단점을 극복하는 것이다. 실험결과는 제안한 방법이 확률적 방법과 질의 위치 이동 방법을 각각 적용한 경우보다 더 나은 성능을 나타낸다는 것을 보여준다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제10권6호
• /
• pp.2527-2545
• /
• 2016

Measuring the similarity of given samples is a key problem of recognition, clustering, retrieval and related applications. A number of works, e.g. kernel method and metric learning, have been contributed to this problem. The challenge of similarity learning is to find a similarity robust to intra-class variance and simultaneously selective to inter-class characteristic. We observed that, the similarity measure can be improved if the data distribution and hidden semantic information are exploited in a more sophisticated way. In this paper, we propose a similarity learning approach for retrieval and recognition. The approach, termed as LDA-FEK, derives free energy kernel (FEK) from Latent Dirichlet Allocation (LDA). First, it trains LDA and constructs kernel using the parameters and variables of the trained model. Then, the unknown kernel parameters are learned by a discriminative learning approach. The main contributions of the proposed method are twofold: (1) the method is computationally efficient and scalable since the parameters in kernel are determined in a staged way; (2) the method exploits data distribution and semantic level hidden information by means of LDA. To evaluate the performance of LDA-FEK, we apply it for image retrieval over two data sets and for text categorization on four popular data sets. The results show the competitive performance of our method.

• 한국정보과학회논문지:데이타베이스
• /
• 제36권5호
• /
• pp.366-373
• /
• 2009

이 논문은 이미지 데이터베이스를 위한 유사도 순위 매김 알고리즘을 제시한다. 이미지 검색의 문제점 중 하나가 이미지로부터 자동적으로 계산한 하위 레벨 특성과 인간 지각과의 의미 차이이며, 검색시에 이미지 유사도 측정을 위해 많은 알고리즘에서는 민코프스키 측정법($L_p$-norm)을 사용하고 있다. 그러나 민코프스키 측정법은 인간 시각 시스템의 비선형적 특성과 문맥 정보를 반영하지 못한다. 본 알고리즘에서는 인간 지각의 비선형성과 문맥 정보를 반영하는 유사도와 탐색 알고리즘을 통해 이 문제를 해결한다. 본 알고리즘을 필기체 숫자 이미지 데이터베이스에 적용하여 성능의 우수성과 효과를 증명하였다.

• 제어로봇시스템학회논문지
• /
• 제8권4호
• /
• pp.333-337
• /
• 2002

A content-based image retrieval method using color, texture, and shape features is proposed in this paper. A region segmentation technique using PIM(Picture Information Measure) entropy is used for similarity indexing. For segmentation, a color image is first transformed to a gray image and it is divided into n$\times$n non-overlapping blocks. Entropy using PIM is obtained from each block. Adequate variance to perform good segmentation of images in the database is obtained heuristically. As variance increases up to some bound, objects within the image can be easily segmented from the background. Therefore, variance is a good indication for adequate image segmentation. For high variance image, the image is segmented into two regions-high and low entropy regions. In high entropy region, hue-saturation-intensity and canny edge histograms are used for image similarity calculation. For image having lower variance is well represented by global texture information. Experiments show that the proposed method displayed similar images at the average of 4th rank for top-10 retrieval case.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2003년도 하계종합학술대회 논문집 Ⅲ
• /
• pp.1515-1518
• /
• 2003

This paper suggests the condensed two-stage retrieval method for fast image retrieval in the content-based image retrieval system, and proves the validity of the performance. The condensed two-stage retrieval method reduces the overall response time remarkably while it maintains relevance with the conventional exhaustive search method. It is explained by properties of the Cauchy-Schwartz inequality. In experimental result, it turns out that there is an optimal value of condensation ratio which minimizes the overall response time. We analyze the optimal condensation ratio by modeling a similarity computation time mathematically.