Multiresolution content-based image retrieval and clustering in large visual databases

To an increasing extent, applications demand the retrieval capability based on image content. As a result, large image database systems need to be built to effectively and efficiently access visual data on the basis of content. In this process, significant features must first be extracted from the image data. Merging results obtained from these heterogeneous features is an important issue. There is the need to consider the semantics of images to improve the retrieval performance, and usually the semantics of images is not represented by individual feature vectors. In this dissertation, we introduce the Advanced Multi-resolution Image Retrieval ( AMIR) system. In the AMIR system, we use a generalized hierarchical image decomposition method based on nona-tree to represent images in visual databases. Texture and color features of images will be extracted. Using the multi-resolution property of wavelet transform, the system extracts texture features at different scales. We propose a neural network based model to merge the results from heterogeneous features. We also present a semantic-based clustering method to classify images based on their semantics and heterogeneous features. Each semantic cluster contains a set of subclusters that are represented by the heterogeneous features that the images contain. We also address clustering very large database. Here we propose a novel approach, WaveCluster , based on wavelet transform that applies signal processing methods on database problems. Comprehensive experimental results show the performance of the proposed approaches.