Active learning in very large databases

Query-by-example and query-by-keyword both suffer from the problem of “aliasing,” meaning that example-images and keywords potentially have variable interpretations or multiple semantics. For discerning which semantic is appropriate for a given query, we have established that combining active learning with kernel methods is a very effective approach. In this work, we first examine active-learning strategies, and then focus on addressing the challenges of two scalability issues: scalability in concept complexity and in dataset size. We present remedies, explain limitations, and discuss future directions that research might take.     

Active learning for image retrieval with Co-SVM

In relevance feedback algorithms, selective sampling is often used to reduce the cost of labeling and explore the unlabeled data. In this paper, we proposed an active learning algorithm, Co-SVM, to improve the performance of selective sampling in image retrieval. In Co-SVM algorithm, color and texture are naturally considered as sufficient and uncorrelated views of an image. SVM classifiers are learned in color and texture feature subspaces, respectively. Then the two classifiers are used to classify the unlabeled data. These unlabeled samples which are differently classified by the two classifiers are chose to label. The experimental results show that the proposed algorithm is beneficial to image retrieval.     

Geometry-based image retrieval in binary image databases

In this paper, a geometry-based image retrieval system is developed for multi-object images. We model both shape and topology of image objects using a structured representation called curvature tree (CT). The hierarchy of the CT reflects the inclusion relationships between the image objects. To facilitate shape-based matching, triangle-area representation (TAR) of each object is stored at the corresponding node in the CT. The similarity between two multi-object images is measured based on the maximum similarity subtree isomorphism (MSSI) between their CTs. For this purpose, we adopt a recursive algorithm to solve the MSSI problem and a very effective dynamic programming algorithm to measure the similarity between the attributed nodes. Our matching scheme agrees with many recent findings in psychology about the human perception of multi-object images. Experiments on a database of 13500 real and synthesized medical images and the MPEG-7 CE-1 database of 1400 shape images have shown the effectiveness of the proposed method.     

Active learning SVM with regularization path for image classification

In classification problems, many different active learning techniques are often adopted to find the most informative samples for labeling in order to save human labors. Among them, active learning support vector machine (SVM) is one of the most representative approaches, in which model parameter is usually set as a fixed default value during the whole learning process. Note that model parameter is closely related to the training set. Hence dynamic parameter is desirable to make a satisfactory learning performance. To target this issue, we proposed a novel algorithm, called active learning SVM with regularization path, which can fit the entire solution path of SVM for every value of model parameters. In this algorithm, we first traced the entire solution path of the current classifier to find a series of candidate model parameters, and then used unlabeled samples to select the best model parameter. Besides, in the initial phase of training, we constructed a training sample sets by using an improved K-medoids cluster algorithm. Experimental results conducted from real-world data sets showed the effectiveness of the proposed algorithm for image classification problems.     

Inductive learning in deductive databases

Most current applications of inductive learning in databases take place in the context of a single extensional relation. The authors place inductive learning in the context of a set of relations defined either extensionally or intentionally in the framework of deductive databases. LINUS, an inductive logic programming system that induces virtual relations from example positive and negative tuples and already defined relations in a deductive database, is presented. Based on the idea of transforming the problem of learning relations to attribute-value form, several attribute-value learning systems are incorporated. As the latter handle noisy data successfully, LINUS is able to learn relations from real-life noisy databases. The use of LINUS for learning virtual relations is illustrated, and a study of its performance on noisy data is presented     

Smart image design for large image databases

We propose an active image information system based upon the concept of smart images. A smart image is an image with an associated knowledge structure, consisting of protocols, hot spots, active indexes and attributes. This active image information system enables us to accomplish the objectives of timely delivery and easy accessibility by handling long-duration operations and supporting unitary image information usage. The experimental prototype of the smart image system is described.     

Active learning for object-based image classification using predefined training objects

Object-based image analysis (OBIA) is a new remote-sensing-based image processing technology that has become popular in recent years. In spite of its remarkable advantages, the segmentation results that it generates feature a large number of mixed objects owing to the limitations of OBIA segmentation technology. The mixed objects directly influence the acquisition of training samples and the labelling of objects and thus affect the stability of classification performance. In light of this issue, this article evaluates the influence of classification uncertainty on classification performance and proposes a sampling strategy based on active learning. This sampling strategy is novel in two ways: (1) information entropy is used to evaluate the classification uncertainty of segmented objects; all segmented objects are classified as having zero or non-zero entropies, and the latter are arranged in terms of decreasing entropy. (2) Based on an evaluation of the influence of classification uncertainty on classification performance, an active learning technology is developed. A certain proportion of zero-entropy objects is acquired via random sampling used as seed training samples for active learning, non-zero-entropy objects are used as a candidate set for active learning, and the entropy query-by-bagging (EQB) algorithm is used to conduct active learning to acquire optimal training samples. In this study, three groups of high-resolution images were tested. The test results show that zero-entropy and non-zero-entropy objects are indispensable to the classifier, where the optimal range of the ratio of combination of the two is between 0.2 and 0.6. Moreover, the proposed sampling strategy can effectively improve the stability and accuracy of classification.     

Multiresolution similarity search in image databases

Typically searching image collections is based on features of the images. In most cases the features are based on the color histogram of the images. Similarity search based on color histograms is very efficient, but the quality of the search results is often rather poor. One of the reasons is that histogram-based systems only support a specific form of global similarity using the whole histogram as one vector. But there is more information in a histogram than the distribution of colors. This paper has two contributions: (1) a new generalized similarity search method based on a wavelet transformation of the color histograms and (2) a new effectiveness measure for image similarity search. Our generalized similarity search method has been developed to allow the user to search for images with similarities on arbitrary detail levels of the color histogram. We show that our new approach is more general and more effective than previous approaches while retaining a competitive performance.     

Information retrieval in document image databases

With the rising popularity and importance of document images as an information source, information retrieval in document image databases has become a growing and challenging problem. In this paper, we propose an approach with the capability of matching partial word images to address two issues in document image retrieval: word spotting and similarity measurement between documents. First, each word image is represented by a primitive string. Then, an inexact string matching technique is utilized to measure the similarity between the two primitive strings generated from two word images. Based on the similarity, we can estimate how a word image is relevant to the other and, thereby, decide whether one is a portion of the other. To deal with various character fonts, we use a primitive string which is tolerant to serif and font differences to represent a word image. Using this technique of inexact string matching, our method is able to successfully handle the problem of heavily touching characters. Experimental results on a variety of document image databases confirm the feasibility, validity, and efficiency of our proposed approach in document image retrieval.     

Similarity searching in medical image databases

We propose a method to handle approximate searching by image content in medical image databases. Image content is represented by attributed relational graphs holding features of objects and relationships between objects. The method relies on the assumption that a fixed number of “labeled” or “expected” objects (e.g., “heart”, “lungs”, etc.) are common in all images of a given application domain in addition to a variable number of “unexpected” or “unlabeled” objects (e.g., “tumor”, “hematoma”, etc.). The method can answer queries by example, such as “find all X-rays that are similar to Smith's X-ray”. The stored images are mapped to points in a multidimensional space and are indexed using state-of-the-art database methods (R-trees). The proposed method has several desirable properties: (a) Database search is approximate, so that all images up to a prespecified degree of similarity (tolerance) are retrieved. (b) It has no “false dismissals” (i.e., all images qualifying query selection criteria are retrieved). (c) It is much faster than sequential scanning for searching in the main memory and on the disk (i.e., by up to an order of magnitude), thus scaling-up well for large databases     

Interactive databases for leisure learning

An interactive database, with the database information provided by an expert, and presented so that it can easily be “questioned”, can be described as a simplified form of expert system.Such mini-expert systems are within the memory capacity of small home microcomputers and can serve as an introduction to microcomputing for inexperienced or even reluctant users. By supplying tailor-made and personalised information and advice, such programs are not only educative but useful, providing one answer to the question “What can I do with a micro besides playing arcade games?”In 1984, Ebury Software began developing a series of practical software titles for launch in 1985. Titles planned included gardening and cookery advice, home decorating and health and beauty care. The choice of subjects reflected our position within the National Magazine Company, whose other imprints include the magazines Good Housekeeping, Cosmopolitan and Harper's and Queen, as well as Ebury Press.Designing such a series so that it would be genuinely educational, practical, flexible and easy for inexperienced micro owners to use required some careful thought. Using the gardening project as a model, the process of our self-education, and the principles we developed as we went along, will be described in the paper.     

Emergent semantics through interaction in image databases

In this paper, we briefly discuss some aspects of image semantics and the role that it plays for the design of image databases. We argue that images don't have an intrinsic meaning, but that they are endowed with a meaning by placing them in the context of other images and by the user interaction. From this observation, we conclude that, in an image, database users should be allowed to manipulate not only the individual images, but also the relation between them. We present an interface model based on the manipulation of configurations of images     

Query processing issues in region-based image databases

Many modern image database systems adopt a region-based paradigm, in which images are segmented into homogeneous regions in order to improve the retrieval accuracy. With respect to the case where images are dealt with as a whole, this leads to some peculiar query processing issues that have not been investigated so far in an integrated way. Thus, it is currently hard to understand how the different alternatives for implementing the region-based image retrieval model might impact on performance. In this paper, we analyze in detail such issues, in particular the type of matching between regions (either one-to-one or many-to-many). Then, we propose a novel ranking model, based on the concept of Skyline, as an alternative to the usual one based on aggregation functions and k-Nearest Neighbors queries. We also discuss how different query types can be efficiently supported. For all the considered scenarios we detail efficient index-based algorithms that are provably correct. Extensive experimental analysis shows, among other things, that: (1) the 1–1 matching type has to be preferred to the N–M one in terms of efficiency, whereas the two have comparable effectiveness, (2) indexing regions rather than images performs much better, and (3) the novel Skyline ranking model is consistently the most efficient one, even if this sometimes comes at the price of a reduced effectiveness.     

Mining spatial association rules in image databases

In this paper, we propose a novel spatial mining algorithm, called 9DLT-Miner, to mine the spatial association rules from an image database, where every image is represented by the 9DLT representation. The proposed method consists of two phases. First, we find all frequent patterns of length one. Next, we use frequent k-patterns (k ? 1) to generate all candidate (k + 1)-patterns. For each candidate pattern generated, we scan the database to count the pattern’s support and check if it is frequent. The steps in the second phase are repeated until no more frequent patterns can be found. Since our proposed algorithm prunes most of impossible candidates, it is more efficient than the Apriori algorithm. The experiment results show that 9DLT-Miner runs 2-5 times faster than the Apriori algorithm.     

An automatic feature generation approach to multiple instance learning and its applications to image databases

Automatic content-based image categorization is a challenging research topic and has many practical applications. Images are usually represented as bags of feature vectors, and the categorization problem is studied in the Multiple-Instance Learning (MIL) framework. In this paper, we propose a novel learning technique which transforms the MIL problem into a standard supervised learning problem by defining a feature vector for each image bag. Specifically, the feature vectors of the image bags are grouped into clusters and each cluster is given a label. Using these labels, each instance of an image bag can be replaced by a corresponding label to obtain a bag of cluster labels. Data mining can then be employed to uncover common label patterns for each image category. These label patterns are converted into bags of feature vectors; and they are used to transform each image bag in the data set into a feature vector such that each vector element is the distance of the image bag to a distinct pattern bag. With this new image representation, standard supervised learning algorithms can be applied to classify the images into the pre-defined categories. Our experimental results demonstrate the superiority of the proposed technique in categorization accuracy as compared to state-of-the-art methods.     

Adaptively browsing image databases with PIBE

Browsing large image collections is a complex and often tedious task, due to the semantic gap existing between the user subjective notion of similarity and the one according to which a browsing system organizes the images. In this paper we propose PIBE, an adaptive image browsing system, which provides users with a hierarchical view of images (the Browsing Tree) that can be customized according to user preferences. A key feature of PIBE is that it maintains local similarity criteria for each portion of the Browsing Tree. This makes it possible both to avoid costly global reorganization upon execution of user actions and, combined with a persistent storage of the Browsing Tree, to efficiently support multiple browsing tasks. We present the basic principles of PIBE and report experimental results showing the effectiveness of its browsing and personalization functionalities.     

Texture-based pattern retrieval from image databases

Recent technological advances have made it possible to process and store large amounts of image data. Perhaps the most impressive example is the accumulation of image data in scientific applications such as medical or satellite imagery. However, in order to realize their full potential, tools for efficient extraction of information and for intelligent searches in image databases need to be developed. This paper describes a new approach to image data retrieval which allows queries to be composed of local intensity patterns. The intensity pattern is converted into a feature representation of reduced dimensionality which can be used for searching similar-looking patterns in the database. This representation is obtained by filtering the pattern with a bank of scale and orientation selective filters modeled using Gabor functions. Experimental results are presented which illustrate that the proposed representation preserves the perceptual similarities, and provides a powerful tool for content-based image retrieval.