Web Image Classification Research Articles

이미지 단어집과 관심영역 자동추출을 사용한 이미지 분류

플리커, 페이스북과 같은 대용량 소셜 미디어 공유 사이트의 발전으로 이미지 정보가 매우 빠르게 증가하고 있다. 이에 따라 소셜 이미지를 정확하게 검색하기 위한 다양한 연구가 활발히 진행되고 있다. 이미지 태그들의 의미적 연관성을 이용하여 태그기반의 이미지 검색의 정확도를 높이고자 하는 연구를 비롯하여 이미지 단어집(Bag of Visual Words)을 기반으로 웹 이미지를 분류하는 연구도 다양하게 진행되고 있다. 본 논문에서는 이미지에서 배경과 같은 중요도가 떨어지는 정보를 제거하여 중요부분을 찾는 GBVS(Graph Based Visual Saliency)모델을 기존 연구에 사용할 것을 제안한다. 제안하는 방법은 첫 번째, 이미지 태그들의 의미적 연관성을 이용해 1차 분류된 데이터베이스에 SIFT알고리즘을 사용하여 이미지 단어집(BoVW)을 만든다. 두 번째, 테스트할 이미지에 GBVS를 통해서 이미지의 관심영역을 선택하여 테스트한다. 의미연관성 태그와 SIFT기반의 이미지 단어집을 사용한 기존의 방법에 GBVS를 적용한 결과 더 높은 정확도를 보임을 확인하였다. As social multimedia sites are getting popular such as Flickr and Facebook, the amount of image information has been increasing very fast. So there have been many studies for accurate social image retrieval. Some of them were web image classification using semantic relations of image tags and BoVW(Bag of Visual Words). In this paper, we propose a method to detect salient region in images using GBVS(Graph Based Visual Saliency) model which can eliminate less important region like a background. First, We construct BoVW based on SIFT algorithm from the database of the preliminary retrieved images with semantically related tags. Second, detect salient region in test images using GBVS model. The result of image classification showed higher accuracy than the previous research. Therefore we expect that our method can classify a variety of images more accurately.

Multi-view semi-supervised web image classification via co-graph

We propose a new multi-view semi-supervised learning method named co-graph for web image classification. Co-graph combines multiple graphs together, each modeling the data points on one view, and enhances learners incrementally with co-training strategy by exploiting unlabeled data. Learners are locally co-trained in co-graph for the enhancement, i.e., only a part of local models in graphs, named dominant local models, need to be updated instead of the total. We also extend the co-graph algorithm to a general framework of local co-training over multiple graphs that is compatible with the common graph-based learning algorithms. Co-graph builds a bridge between graph-based methods and co-training, and contains the double label propagation: one propagates labels from labeled data to unlabeled data in each single view, and the other exchanges high-confidence label information across different views. Experimental results demonstrate the effectiveness of co-graph in web image classification.

Web page and image semi-supervised classification with heterogeneous information fusion

Web data, such as web pages and web images, can be naturally partitioned into multiple heterogeneous attribute sets. Concretely speaking, web pages consist of hyperlink and contents, and web images consist of the textual and visual information. In this paper, we propose a new multi-view semi-supervised learning method, named local co-training, for web page and image classification. Local co-training employs local linear models to represent data points on each view (i.e. one attribute set), and iteratively refines them using unlabelled data with co-training strategy. In each iteration, only a part of local models that we call dominant local models needs to be incrementally updated. The method is thus efficient and fit for the learning of large-scale web data. In addition, we introduce a new measurement based on both the confidence and the disagreement to describe which unlabelled examples are ‘good’ for the enrichment of training sets. Local co-training builds a bridge between two dominant types of semi-supervised methods: graph-based methods and co-training. Experiments on web page and web image datasets demonstrate that local co-training can effectively improve the classification performance by exploiting multiple attribute sets and unlabelled data.

Using Hilbert scan on statistical color space partitioning

This study proposes a method to combine the k-Nearest Neighbor (k-NN) algorithm and the Support Vector Machine (SVM) method to increase the image annotation accuracy. Image annotation is widely employed in domains such as web image classification, search, military, and biomedicine. Although the traditional Border/Interior pixel Classification (BIC) features are very efficient and compact when applied to image annotation to capture color, shape, and texture information, the color space histogram utilization rates are not balanced. The experiment results show that the Hilbert-scan method and the One-pass Partitioning Method (OPM) can effectively overcome the imbalance problem.

Web Image Classification Using an Optimized Feature Set

Redundant images currently abundant in World Wide Web pages need to be removed in order to transform or simplify the Web pages for suitable display in small-screened devices. Classifying removable images on the Web pages according to their uniqueness of content will allow simpler representation of Web pages. For such classification, machine learning based methods can be used to categorize images into two groups; eliminable and non-eliminable. We use two representative learning methods, the Naïve Bayesian classifier and C4.5 decision trees. For our Web image classification, we propose new features that have expressive power for Web images to be classified. We apply image samples to the two classifiers and analyze the results. In addition, we propose an algorithm to construct an optimized subset from a whole feature set, which includes most influential features for the purposes of classification. By using the optimized feature set, the accuracy of classification is found to improve markedly.

기계학습 기반의 웹 이미지 분류

Although image occupies a large part of importance on the Web documents, there have not been many researches for analyzing and understanding it. Many Web images are used for carrying important information but others are not used for it. In this paper classify the Web images from presently served Web sites to erasable or non-erasable classes. based on machine learning methods. For this research, we have detected 16 special and rich features for Web images and experimented by using the Baysian and decision tree methods. As the results, F-measures of 87.09%, 82.72% were achived for each method and particularly, from the experiments to compare the effects of feature groups, it has proved that the added features on this study are very useful for Web image classification.