Research On Image-based Cultural Relic Retrieval And Ontology-based Annotation

With the rapid development of Content-based image retrieval and Ontology techniques, image-based cultural relic retrieval and ontology-based annotation has became an important and challenging research topic, which will change the traditional cultural relic data management method. In this dissertation, some key techniques of image-based cultural relic retrieval and annotation, such as local features extraction, image semantic acquirment, cultural relic ontology construction and semantic annotation, etc, has been explored and studied. The research work is valuable both in theory and application. The main work and innovations are listed as follows:(1) The local feature extraction technique, which is a novel breakthrough of the existing content-based image retrieval framework, is firstly introduced into the culture image retrieval field, and meanwhile a culture relic image retrieval framwork based on Scale Invariant Feature Transform (SIFT) is proposed. The experimental result on cultural relic image database shows that SIFT feature is not only invariant to rotation and scale variation but also has good retrieval performance on cultural relic images with complex edge. The SIFT-based method improves the performance of our cultural relic image retrieval system.(2) Multiple instance learning (MIL) label method is considered to be a new learning method, which can build the mapping from image's low-level features to its high-level semantic features and obtain image's semantics. Aiming at the problems of bag structure information loss and poor classification performance, when existing neural network algorithm is applied to predict unknown bag label in MIL, a new MIL classification algorithm named RK_BP is presented. First, rough set method is used to reduce the redundant information in the instance feature space. Then K-means algorithm is adopted to cluster and generate a new bag space. Finally back propagation (BP) algorithm is used to classify bags in the new space. Experiments on benchmark data sets and cultural relic image sets show that RK_BP algorithm provides better classification results.(3) Aiming at the ambiguity problem of MIL in images retrieval, a novel image retrieval algorithm named KP-MIL is presented. First, K-means clustering algorithm is used to the instances in two sets, one of which is composed of instances in positive bags and the other is composed of instances in negative bags, so as to find potential positive instances and feature data of bag structure. Then their respective similarities are measured by Radial Basis Function (RBF), and an alpha coefficient is introduced in bag similarity measure as the trade-off between the two similarities. Finally Support Vector Machine (SVM) is used to classify and retrieval. Experiments on SIVAL data sets and cultural relic image sets show that KP-MIL algorithm is good at finding global optimal solution and its performance is superior to other algorithms based on Diverse Density (DD) and structural MIL(4) Focusing on the hierarchy complexity problem of cultural relic classification, a category knowledge base of cultural relic usable by computer is established. A domain ontology model named PYCO is also proposed in our work by extracting metadata model of the category knowledge hierarchy based on ontology techniques. The PYCO can help desiging standardized statement for the terms of pottery yong, possibly analysis the domian knowledge and also establish the relationships between the rules of semantic concepts and their relationship.(5) A method of PYCO-based automatic semantic annotation on images of pottery yong is presented. First Jseg algorithm is used to segment images, and low-level features, which combine with global visual features and local SIFT features, are extracted to form a feature vector of segmented region. Then multi-SVM is adopted to establish the association between low-level features and concepts of attributes. Finally by using the inference rules of semantic concepts of PYCO, automatic annotation of high-level semantic concepts of pottery yong images is implemented. This method effectively supports the ontology-based semantic retrieval of cultural relics.