Jianjiang Lu

Dynamic Adaboost learning with feature selection based on parallel genetic algorithm for image annotation

Image annotation can be formulated as a classification problem. Recently, Adaboost learning with feature selection has been used for creating an accurate ensemble classifier. We propose dynamic Adaboost learning with feature selection based on parallel genetic algorithm for image annotation in MPEG-7 standard. In each iteration of Adaboost learning, genetic algorithm (GA) is used to dynamically generate and optimize a set of feature subsets on which the weak classifiers are constructed, so that an ensemble member is selected. We investigate two methods of GA feature selection: a binary-coded chromosome GA feature selection method used to perform optimal feature subset selection, and a bi-coded chromosome GA feature selection method used to perform optimal-weighted feature subset selection, i.e. simultaneously perform optimal feature subset selection and corresponding optimal weight subset selection. To improve the computational efficiency of our approach, master-slave GA, a parallel program of GA, is implemented. k-nearest neighbor classifier is used as the base classifier. The experiments are performed over 2000 classified Corel images to validate the performance of the approaches.

A novel approach to semantic annotation based on multi-ontologies

Semantic annotation is very crucial to the semantic Web. Most traditional researches just commit the Web resources to a single ontology. However, many practical semantic annotation cases need multi-ontologies. For the sake of solving the semantic annotation problem based on multi-ontologies, one prevalent approach is extending current ontology or integrating multi-ontologies, but both methods are complex problems which have no good solutions till now. Another approach used distributed description logic to denote multi-ontologies and the subsumption relations between concepts in different ontology with a simple bridge rule, but this approach could not describe more complex relations between multi-ontologies. In this paper, a novel approach is proposed to deal with the problem, and we employ a bridge ontology to express the complex relationships between multi-ontologies. The bridge ontology is a peculiar ontology, which can be created and maintained easily, but effective in semantic annotation applications based on multi-ontologies. The approach has the characteristics of low-cost, scalability, robust in the Web circumstance, avoiding the unnecessary ontology extending and integration, and promoting ontology reuse. A semantic annotation framework and an algorithm are proposed As well.

Technique of Image Retrieval Based on Multi-label Image Annotation

In this paper, we propose a novel multi-label image annotation for image retrieval based on annotated keywords. For multi-label image annotation, a bi-coded genetic algorithm is employed to select optimal feature subsets and corresponding optimal weights for every one vs. one SVM classifiers. After an unlabelled image is segmented into several regions with image segmentation algorithm, pre-trained SVMs are used to annotate each region, final label is obtained by merging all the region labels. A novel annotation refinement approach based on PageRank is proposed to get rid of irrelevant labels. Based on multi-label of image, image retrieval system provides keyword-based image retrieval service. Multi-labels can provide abundant descriptions for image content in semantic level, and experiment results shows the multi-label annotation algorithm can improve precision and recall of image retrieval.

Reasoning within extended fuzzy description logic

An extended fuzzy description logic is proposed to increase expressive power for complex fuzzy information. We introduce cut sets of the fuzzy concepts and fuzzy roles as atomic concepts and atomic roles, and inherit concept constructors of the description logics to support a new logic system for fuzzy knowledge representation. We define the syntax, semantics and knowledge base of the extended fuzzy description logic, and focus on sat-domain and consistency as main reasoning tasks. We also present sound and complete tableau algorithms for these reasoning tasks, and prove the complexity of them is PSPACE-complete. Compared with the other fuzzy description logics, the extended fuzzy description logic can express more wide fuzzy information.

Feature Selection Based on Genetic Algorithm for CBIR

Automated techniques to optimize feature descriptor weights and select optimum feature descriptor subset are desirable as a way to enhance the performance of content based image retrieval system. In our system, all the MPEG-7 image feature descriptors including color descriptors, texture descriptors and shape descriptors are used to represent low-level image features. We use a real coded chromosome genetic algorithm (GA) and k-nearest neighbor (k-NN) classification accuracy as fitness function to optimize weights. Meanwhile, a binary one and k-NN classification accuracy combining with the size of feature descriptor subset as fitness function are used to select optimum feature descriptor subset. Furthermore, we propose two kinds of two-stage feature selection schemes for weight optimization and descriptor subset selection, which are the integration of a real coded GA and a binary one. The experimental results over 2000 classified Corel images show that with weight optimization, the accuracy of image retrieval system is improved; with the selection of optimum feature descriptor subset, both the accuracy and the efficiency are improved.

A Framework for Moving Target Detection, Recognition and Tracking in UAV Videos

In this paper, we present a compound framework for moving target detection, recognition and tracking based on different altitude UAV-captured videos. The novel idea of “Divide and Merge” included in our framework is expressed as follows. Firstly, we detect the small and slow moving targets using forward-backward MHI. Secondly, two distinct tracking algorithms, Particle Filter and Mean Shift, are applied to track moving targets in different altitude UAV-captured videos. Then, recognition module divides into two classes: instance recognition and category recognition. The former identifies the target, which is occluded by trees or buildings and reappears later, and the latter classifies the detected target into one category by HoG-based SVM classifier. Besides, recognition-based abnormal target detection and clustering-based abnormal trajectory detection are added to our framework. Armed with this framework, the moving targets can be tracked in real-time and the recognized target or abnormal trajectory gives the alarm in seconds.

Distributed Reasoning with Fuzzy Description Logics

By the development of Semantic Web, increasing demands for vague and distributed information representation have triggered a mass of theoretical and applied researches of fuzzy and distributed ontologies, whose main logical infrastructures are fuzzy and distributed description logics. However, current solutions are proposed respectively on one of these two aspects. By integrating

A saliency detection model using shearlet transform

\mathcal{E}

A Framework of CBIR System Based on Relevance Feedback

-connection into fuzzy description logics, this paper proposes a novel logical approach to couple both fuzzy and distributed features within description logics. The main contribution of these paper is to propose a discrete tableau algorithm to achieve reasoning within this new logical system.

Video Analysis and Trajectory Based Video Annotation System

Visual attention is a mechanism to derive possible locations of objects or regions from natural scenes, and many studies have tried to simulate this mechanism to build saliency detection models, which would accelerate the course of many applications, such as object location, detection and recognition, image segmentation, retrieval and so on. Recently, researchers have tried building the detection models in transform domains. In this paper, a novel saliency detection model using shearlet transform is presented. Firstly, multi-scale feature maps are created. The feature maps built on scaling coefficients are used to generate potential salient regions, which is further used to update the feature maps generated on shearlet coefficients. As these feature maps represent the details of image in multi scale, based on them global and local contrast is calculated to form global and local saliency maps. That is the proposed model obtains the global saliency based on global probability density distribution, and measures the local saliency by calculating the entropy of local areas. By combining the local and global saliency maps, the final saliency maps are obtained. The work of this paper is absolutely a new try to detect saliency regions in shearlet domain, and experimental results demonstrate the saliency detection performance of the novel proposed model.