• Journal of KIISE:Databases
• /
• v.33 no.4
• /
• pp.372-383
• /
• 2006

Among the approaches to image database management, content-based image retrieval (CBIR) is viewed as having the best support for effective searching and browsing of large digital image libraries. Typical CBIR systems allow a user to provide a query image, from which low-level features are extracted and used to find 'similar' images in a database. However, there exists the semantic gap between human visual perception and low-level representations. An effective methodology for overcoming this semantic gap involves relevance feedback to perform feature re-weighting. Current approaches to feature re-weighting require the number of components for a feature representation to be the same for every image in consideration. Following this assumption, they map each component to an axis in the n-dimensional space, which we call the metric space; likewise the feature representation is stored in a fixed-length vector. However, with the emergence of features that do not have a fixed number of components in their representation, existing feature re-weighting approaches are invalidated. In this paper we propose a feature re-weighting technique that supports features regardless of whether or not they can be mapped into a metric space. Our approach analyses the feature distances calculated between the query image and the images in the database. Two-sided confidence intervals are used with the distances to obtain the information for feature re-weighting. There is no restriction on how the distances are calculated for each feature. This provides freedom for how feature representations are structured, i.e. there is no requirement for features to be represented in fixed-length vectors or metric space. Our experimental results show the effectiveness of our approach and in a comparison with other work, we can see how it outperforms previous work.

• IEIE Transactions on Smart Processing and Computing
• /
• v.6 no.4
• /
• pp.262-268
• /
• 2017

Intelligent video surveillance systems have been developed to monitor global areas and find specific target objects using a large-scale database. However, person re-identification presents some challenges, such as pose change and occlusions. To solve the problems, this paper presents an improved person re-identification method using sparse representation and saliency-based dictionary construction. The proposed method consists of three parts: i) feature description based on salient colors and textures for dictionary elements, ii) orthogonal atom selection using cosine similarity to deal with pose and viewpoint change, and iii) measurement of reconstruction error to rank the gallery corresponding a probe object. The proposed method provides good performance, since robust descriptors used as a dictionary atom are generated by weighting some salient features, and dictionary atoms are selected by reducing excessive redundancy causing low accuracy. Therefore, the proposed method can be applied in a large scale-database surveillance system to search for a specific object.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2012.04a
• /
• pp.382-385
• /
• 2012

이 논문에서는 영상 검색(image retrieval) 및 영상 부류(image categorization)을 위하여 영상을 기술할 때 영상의 클레스(class)별로 서로 다른 주요 특징량(feature)에 가중치 를 주는 방법론을 제안한다. 기존에 연구되어온 영상의 특징량 벡터에 가중치를 주는 방식은 모든 영상 클레스에 대하여 동일하게 가중치를 적용하기 때문에 영상이 클레스별로 서로 다른 특징량이 중요하다는 성질을 이용할 수 없다. 영상이 클레 별로 서로 다른 특징량이 중요하다는 성질을 이용하기 위하여 영상의 클레스별로 특징량 벡터에 서로 다른 가중치 벡터(weight vector)를 학습하였다. 그 후 질의 영상(query image)이 입력되면, 기존의 영상 검색 프레임워크(framework)를 통해 데이터베이 스(database)로 부터 미리 정의된 서브 클레스(sub-class)의 수에 해당하는 영상부 집합(subset)을 만들었다. 그리고 영상부 집합의 특징량 벡터들에 클레스별로 각각 학습된 가중치 벡터를 적용하여 특징량 벡터들 간의 거리를 다시 계산하여 리랭킹(re-ranking)하였다. 이 방법론을 UKBench Dataset에 적용하여 실험을 해보았으며 가중치를 주기 전과 비교 하였을 때 더 높은 정확도를 보였다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.12
• /
• pp.3383-3397
• /
• 2023

Scene graphs serve as semantic abstractions of images and play a crucial role in enhancing visual comprehension and reasoning. However, the performance of Scene Graph Generation is often compromised when working with biased data in real-world situations. While many existing systems focus on a single stage of learning for both feature extraction and classification, some employ Class-Balancing strategies, such as Re-weighting, Data Resampling, and Transfer Learning from head to tail. In this paper, we propose a novel approach that decouples the feature extraction and classification phases of the scene graph generation process. For feature extraction, we leverage a transformer-based architecture and design an adaptive calibration function specifically for predicate classification. This function enables us to dynamically adjust the classification scores for each predicate category. Additionally, we introduce a Distribution Alignment technique that effectively balances the class distribution after the feature extraction phase reaches a stable state, thereby facilitating the retraining of the classification head. Importantly, our Distribution Alignment strategy is model-independent and does not require additional supervision, making it applicable to a wide range of SGG models. Using the scene graph diagnostic toolkit on Visual Genome and several popular models, we achieved significant improvements over the previous state-of-the-art methods with our model. Compared to the TDE model, our model improved mR@100 by 70.5% for PredCls, by 84.0% for SGCls, and by 97.6% for SGDet tasks.

• The Journal of the Acoustical Society of Korea
• /
• v.21 no.6
• /
• pp.552-558
• /
• 2002

Recently, based on the human speech recognition (HSR) model of Fletcher, the multi-band speech recognition has been intensively studied by many researchers. As a new automatic speech recognition (ASR) technique, the multi-band speech recognition splits the frequency domain into several sub-bands and recognizes each sub-band independently. The likelihood scores of sub-bands are weighted according to reliabilities of sub-bands and re-combined to make a final decision. This approach is known to be robust under noisy environments. When the noise is stationary a sub-band SNR can be estimated using the noise information in non-speech interval. However, if the noise is non-stationary it is not feasible to obtain the sub-band SNR. This paper proposes the inverse sub-band distance (ISD) weighting, where a distance of each sub-band is calculated by a stochastic matching of input feature vectors and hidden Markov models. The inverse distance is used as a sub-band weight. Experiments on 1500∼1800㎐ band-limited white noise and classical guitar sound revealed that the proposed method could represent the sub-band reliability effectively and improve the performance under both stationary and non-stationary band-limited noise environments.