The objective of the present work is to propose a novel method to extract a stable feature set representative of image content. Each image is represented by a linear combination of fractal orthonormal basis vectors. The mapping coefficients of an image projected onto each orthonormal basis constitute the feature vector. The set of orthonormal basis vectors are generated by utilizing fractal iterative function through target and domain blocks mapping. The distance measure remains consistent, i.e., isometric embedded, between any image pairs before and after the projection onto orthonormal axes. Not only similar images generate points close to each other in the feature space, but also dissimilar ones produce feature points far apart. The above statements are logically equivalent to that distant feature points are guaranteed to map to images with dissimilar contents, while close feature points correspond to similar images. Therefore, utilizing coefficients derived from the proposed linear combination of fractal orthonormal basis as key to search image database will retrieve similar images, while at the same time exclude dissimilar ones. The coefficients associated with each image can be later used to reconstruct the original. The content-based query is performed in the compressed domain. This approach is efficient for content-based query. Scaling, rotational, translation, mirroring and horizontal/vertical flipping variations of a query image are also supported. A symbolic image database system is a system in which a large amount of image data and their related information are represented by both symbolic images and physical images. How to perceive spatial relationships among the components in a symbolic image is an important criterion to find a match between the symbolic image of the scene object and the one being store as a modal in the symbolic image database. Spatial reasoning techniques have been applied to pictorial database, in particular those using 2D strings as an index representation have been successful. In most of the previous approaches for iconic indexing, for simplifying the concerns, they apply the MBR (Minimum bounding rectangle) of two objects to define the spatial relationship between them. Multiple instance learning algorithms provide ways for computer program to improve automatically with experience. Most images are inherently ambiguous disseminators of information. Unfortunately, interfaces to image databases normally involve the user giving the system ambiguous queries. By treating each query as a Multiple-Instance example, we make the ambiguity in each image explicit. In addition, by receiving several positive and negative examples, the system can learn what the user desires. Using the learned concept, the system returns images from the database that are close to that concept. In this project, we propose to apply the Multiple-Instance learning model by deriving the projection vector of fractal orthonormal bases for a small number of training images to learn what images from database are of interest to the user.

With the advent of multimedia computer, the voice and images could be stored in database. How to retrieve the information user want is a heard question. To query the large numbers of digital images which human desired is not a simple task. The studies of traditional image database retrieval use color, shape, and content to analyze a digital image, and create the index file. But they cannot promise that use the similar index files will find the similar images, and the similar images can get the similar index files. In this thesis, we propose a new method to analyze a digital image by fractal code. Fractal coding is an effective method to compress digital image. In fractal code, the image is partitioned into a set of non-overlapping range blocks, and a set of overlapping domain blocks is chosen from the same image. For all range blocks, we need to find one domain block and one iteration function such that the mapping from the domain block is similar to the range block. Two similar images have similar iterated functions, and two similar iterated functions have similar attractors. In these two reasons, we use the iteration function to create index file. We have proved fractal code can be a good index file in chapter 2. In chapter 3, we implement the fractal-based image database. In this system, we used fractal code to create index file, and used Fisher discriminate function, color, complexity, and illumination to decide the output order.

Thesis (Ph. D.)--University of Washington, 2001. / Vita. Includes bibliographical references (leaves 245-272).

The user perspective of image search remains poorly understood. the purpose of this study is to identify and investigate the key issues relevant to a user’s interaction with images and the user’s approach to image search. a deeper understanding of these issues will serve to inform the design of image retrieval systems and in turn better serve the user. Previous research explores areas of information seeking behavior, representation in information science, query formulation, and image retrieval. the theoretical framework for this study includes an articulation of image search scenarios as adapted from Yoon and O’Connor’s taxonomy of image query types, Copeland’s Engineering Design Approach for rigorous qualitative research, and Anderson’s Functional Ontology Construction Model for building robust models of human behavior. a series of semi-structured interviews were conducted with expert-level image users. Interviewees discussed their motivations for image search, types of image searches they pursue, and varied approaches to image search, as well as how they decide that an information need has been met and which factors influence their experience of search. a content analysis revealed themes repeated across responses, including a collection of 23 emergent concepts and 6 emergent categories. a functional analysis revealed further insight into these themes. Results from both analyses may be used as a framework for future exploration of this topic. Implications are discussed and future research directions are indicated. Among possibilities for future research are investigations into collaborative search and ubiquitous image search.

No description available.

The objective of this thesis is to implement a web application designed for recommendation of stock photos. The application gets the input from newspaper articles in Czech or English and, based on the text itself, suggests appropriate stock photos. The implemented application also searches images according to visual similarity. The thesis deals with theoretical aspects of keywords extraction and language of text detection. Further it analyzes possibilities of efficient search for similar vectors that are used in the search component for visually similar images. It also describes the possibilities in development of modern web frontend and backend. The quality of algorithm for recommending stock photos is tested on users. Powered by TCPDF (www.tcpdf.org)

none

This thesis is concerned with improving the effectiveness of nearest neighbour search. Nearest neighbour search is the problem of finding the most similar data-points to a query in a database, and is a fundamental operation that has found wide applicability in many fields. In this thesis the focus is placed on hashing-based approximate nearest neighbour search methods that generate similar binary hashcodes for similar data-points. These hashcodes can be used as the indices into the buckets of hashtables for fast search. This work explores how the quality of search can be improved by learning task specific binary hashcodes. The generation of a binary hashcode comprises two main steps carried out sequentially: projection of the image feature vector onto the normal vectors of a set of hyperplanes partitioning the input feature space followed by a quantisation operation that uses a single threshold to binarise the resulting projections to obtain the hashcodes. The degree to which these operations preserve the relative distances between the datapoints in the input feature space has a direct influence on the effectiveness of using the resulting hashcodes for nearest neighbour search. In this thesis I argue that the retrieval effectiveness of existing hashing-based nearest neighbour search methods can be increased by learning the thresholds and hyperplanes based on the distribution of the input data. The first contribution is a model for learning multiple quantisation thresholds. I demonstrate that the best threshold positioning is projection specific and introduce a novel clustering algorithm for threshold optimisation. The second contribution extends this algorithm by learning the optimal allocation of quantisation thresholds per hyperplane. In doing so I argue that some hyperplanes are naturally more effective than others at capturing the distribution of the data and should therefore attract a greater allocation of quantisation thresholds. The third contribution focuses on the complementary problem of learning the hashing hyperplanes. I introduce a multi-step iterative model that, in the first step, regularises the hashcodes over a data-point adjacency graph, which encourages similar data-points to be assigned similar hashcodes. In the second step, binary classifiers are learnt to separate opposing bits with maximum margin. This algorithm is extended to learn hyperplanes that can generate similar hashcodes for similar data-points in two different feature spaces (e.g. text and images). Individually the performance of these algorithms is often superior to competitive baselines. I unify my contributions by demonstrating that learning hyperplanes and thresholds as part of the same model can yield an additive increase in retrieval effectiveness.

Content-Based Image Retrieval has been an active research area for decades. In a CBIR system, one or more images are used as query to search for similar images. The similarity is measured on the low level features, such as color, shape, edge, texture. First, each image is processed and visual features are extract. Therefore each image becomes a point in the feature space. Then, if two images are close to each other in the feature space, they are considered similar. That is, the k nearest neighbors are considered the most similar images to the query image. In this K-Nearest Neighbor (k-NN) model, semantically similar images are assumed to be clustered together in a single neighborhood in the high-dimensional feature space. Unfortunately semantically similar images with different appearances are often clustered into distinct neighborhoods, which might scatter in the feature space. Hence, confinement of the search results to a single neighborhood is the latent reason of the low recall rate of typical nearest neighbor techniques. In this dissertation, a new image retrieval technique - the Query Decomposition (QD) model is introduced. QD facilitates retrieval of semantically similar images from multiple neighborhoods in the feature space and hence bridges the semantic gap between the images’ low-level feature and the high-level semantic meaning. In the QD model, a query may be decomposed into multiple subqueries based on the user’s relevance feedback to cover multiple image clusters which contain semantically similar images. The retrieval results are the k most similar images from multiple discontinuous relevant clusters. To apply the benifit from QD study, a mobile client-side relevance feedback study was conducted. With the proliferation of handheld devices, the demand of multimedia information retrieval on mobile devices has attracted more attention. A relevance feedback information retrieval process usually includes several rounds of query refinement. Each round incurs exchange of tens of images between the mobile device and the server. With limited wireless bandwidth, this process can incur substantial delay making the system unfriendly iii to use. The Relevance Feedback Support (RFS) structure that was designed in QD technique was adopted for Client-side Relevance Feedback (CRF). Since relevance feedback is done on client side, system response is instantaneous significantly enhancing system usability. Furthermore, since the server is not involved in relevance feedback processing, it is able to support thousands more users simultaneously. As the QD technique improves on the accuracy of CBIR systems, another study, which is called In-Memory relevance feedback is studied in this dissertation. In the study, we improved the efficiency of the CBIR systems. Current methods rely on searching the database, stored on disks, in each round of relevance feedback. This strategy incurs long delay making relevance feedback less friendly to the user, especially for very large databases. Thus, scalability is a limitation of existing solutions. The proposed in-memory relevance feedback technique substantially reduce the delay associated with feedback processing, and therefore improve system usability. A data-independent dimensionality-reduction technique is used to compress the metadata to build a small in-memory database to support relevance feedback operations with minimal disk accesses. The performance of this approach is compared with conventional relevance feedback techniques in terms of computation efficiency and retrieval accuracy. The results indicate that the new technique substantially reduces response time for user feedback while maintaining the quality of the retrieval. In the previous studies, the QD technique relies on a pre-defined Relevance Support Support structure. As the result and user experience indicated that the structure might confine the search range and affect the result. In this dissertation, a novel Multiple Direction Search framework for semi-automatic annotation propagation is studied. In this system, the user interacts with the system to provide example images and the corresponding annotations during the annotation propagation process. In each iteration, the example images are dynamically clustered and the corresponding annotations are propagated separately to each cluster: images in the local neighborhood are annotated. Furthermore, some of those images are returned to the user for further annotation. As the user marks more images, iv the annotation process goes into multiple directions in the feature space. The query movements can be treated as multiple path navigation. Each path could be further split based on the user’s input. In this manner, the system provides accurate annotation assistance to the user - images with the same semantic meaning but different visual characteristics can be handled effectively. From comprehensive experiments on Corel and U. of Washington image databases, the proposed technique shows accuracy and efficiency on annotating image databases.

No description available.