Daniel S. Kaster

PRoSPer: Perceptual similarity queries in medical CBIR systems through user profiles

In this paper, we present a novel approach to perform similarity queries over medical images, maintaining the semantics of a given query posted by the user. Content-based image retrieval systems relying on relevance feedback techniques usually request the users to label relevant/irrelevant images. Thus, we present a highly effective strategy to survey user profiles, taking advantage of such labeling to implicitly gather the user perceptual similarity. The profiles maintain the settings desired for each user, allowing tuning of the similarity assessment, which encompasses the dynamic change of the distance function employed through an interactive process. Experiments on medical images show that the method is effective and can improve the decision making process during analysis.

Unsupervised scaling of multi-descriptor similarity functions for medical image datasets

Content-based search has proven to be a proper complement to textual queries over medical image databases. In many applications, employing multiple image descriptors and combining the respective distance functions using adequate scale factors improves the retrieval accuracy. However, the existing weighting methods are either exhaustive or supervised. In this paper, we present the Fractal-scaled Product Metric, an unsupervised method to determine a scale factor among features in multi-descriptor image similarity assessment based on the Fractal Theory. The composite distance function obtained is not limited to dimensional image descriptors and enables using scalable indexing structures. Experiments have shown that the proposed method determines near-optimal scale factors for the descriptors involved, and always improves the precision of the results, outperforming the individual descriptors up to 31% on the average precision.

Improving the Pruning Ability of Dynamic Metric Access Methods with Local Additional Pivots and Anticipation of Information

Metric Access Methods (MAMs) have been proved to allow performing similarity queries over complex data more efficiently than other access methods. They can be considered dynamic or static depending on the pivot type used in their construction. Global pivots tend to compromise the dynamicity of MAMs, as eventual pivot-related updates must be propagated through the entire structure, while local pivots allow this maintenance to occur locally. Several applications handle online complex data and, consequently, demand efficient dynamic indexes to be successful. In this context, this work presents two techniques for improving the pruning ability of dynamic MAMs: (i) using cutting local additional pivots to reduce distance calculations and (ii) anticipating information from child nodes to reduce unnecessary disk accesses. The experiments reveal significant improvements in a dynamic MAM, reducing execution time in more than 50 % for similarity queries posed on datasets ranging from moderate to high dimensionality and cardinality.

Efficient Execution of Conjunctive Complex Queries on Big Multimedia Databases

This paper proposes an approach to efficiently execute conjunctive queries on big complex data together with their related conventional data. The basic idea is to horizontally fragment the database according to criteria frequently used in query predicates. The collection of fragments is indexed to efficiently find the fragment(s) whose contents satisfy some query predicate(s). The contents of each fragment are then indexed as well, to support efficient filtering of the fragment data according to other query predicate(s) conjunctively connected to the former. This strategy has been applied to a collection of more than 106 million images together with their related conventional data. Experimental results show considerable performance gain of the proposed approach for queries with conventional and similarity-based predicates, compared to the use of a unique metric index for the entire database contents.

Improving Metric Access Methods with Bucket Files

Modern applications deal with complex data, where retrieval by similarity plays an important role in most of them. Complex data whose primary comparison mechanisms are similarity predicates are usually immersed in metric spaces. Metric Access Methods MAMs exploit the metric space properties to divide the metric space into regions and conquer efficiency on the processing of similarity queries, like range and k-nearest neighbor queries. Existing MAM use homogeneous data structures to improve query execution, pursuing the same techniques employed by traditional methods developed to retrieve scalar and multidimensional data. In this paper, we combine hashing and hierarchical ball partitioning approaches to achieve a hybrid index that is tuned to improve similarity queries targeting complex data sets, with search algorithms that reduce total execution time by aggressively reducing the number of distance calculations. We applied our technique in the Slim-tree and performed experiments over real data sets showing that the proposed technique is able to reduce the execution time of both range and k-nearest queries to at least half of the Slim-tree. Moreover, this technique is general to be applied over many existing MAM.

Using Visual Analysis to Weight Multiple Signatures to Discriminate Complex Data

Complex data is usually represented through signatures, which are sets of features describing the data content. Several kinds of complex data allow extracting different signatures from an object, representing complementary data characteristics. However, there is no ground truth of how balancing these signatures to reach an ideal similarity distribution. It depends on the analyst intent, that is, according to the job he/she is performing, a few signatures should have more impact in the data distribution than others. This work presents a new technique, called Visual Signature Weighting (ViSW), which allows interactively analyzing the impact of each signature in the similarity of complex data represented through multiple signatures. Our method provides means to explore the tradeoff of prioritizing signatures over the others, by dynamically changing their weight relation. We also present case studies showing that the technique is useful for global dataset analysis as well as for inspecting subspaces of interest.

TendeR-Sims - Similarity Retrieval System for Public Tenders.

TendeR-Sims (Tender Retrieval by Similarity) is a system that helps to search for satisfiable request for tender’s lots in a database by filtering irrelevant lots, so companies can easily discover the contracts they can win. The system implements the Similarity-aware Relational Division Operator in a commercial Relational Database Management System (RDBMS), and compares products by combining a path distance in a preprocessed ontology with a textual distance. Tender-Sims focuses on answering the following query: select the lots where a company has a similar enough item for each of all required items. We evaluated our proposed system employing a dataset composed of product catologs of Brazilian companies in the food market and real requests for tenders with known results. In the presented experiments, TendeR Sims achieved up to 66% cost reduction at 90% recall when compared to the ground truth.

One index to dominate them all: domain indexes for improving queries across multiple tables

Relational Database Management Systems (RDBMS) organize data into relations, representing them as tables. Most of queries executed over them are optimized by index structures. However, considering queries that require scanning indexes across multiple tables, the common approach involves scanning multiple indexes and combining their results, which is potentially costly, especially regarding similarity queries over complex data. This paper proposes a new type of index for modern RDBMS called domain index. Such proposal consists of indexes that allow searching columns of the same type, across multiple tables, with a single index scan, hence with superior performance. To evaluate our proposal, we carried out experiments (i) over a medical image dataset, to evaluate the performance in content-based similarity queries; and (ii) over a flow-based intrusion detection dataset, to evaluate the performance in conventional queries both in a real scenario and over synthetic data so to evaluate scalability. The results exhibit the higher performance of domain indexes. Specifically, the gains reached up to 42.9+ in similarity queries and up to 65.9+ in conventional queries. As the first paper on this subject, we expect this work to provide a basis for further developments on indexing techniques over domains of attributes within modern RDBMS.

A Similarity-Based Approach for Financial Time Series Analysis and Forecasting

Financial time series analysis have been attracting research interest for several years. Many works have been proposed to perform economic series forecasting, however, it still is a hard endeavor to develop a general model that is able to handle the chaotic nature of the markets. Artificial intelligence methods such as artificial neural networks and support vector machines arose as promising alternatives, but they hide the processing semantics, limiting the result interpretation. In addition, one of the main drawbacks of the existing solutions is that they usually cannot be easily employed as building blocks of new analysis tools. This paper presents a new approach to financial time series forecasting based on similarity between series patterns using a database-driven architecture. We propose a new feature extractor based on visual features associated with a boosted instance-based learning classifier to predict a share’s behavior, thus improving the human analyst understanding and validation of the results. The analysis is defined through extended SQL instructions and executed over a fast and scalable engine, which makes our solution adequate to provide data analysis support for new applications handling large time series datasets. We also present experiments performed on data obtained from distinct market shares. The achieved results show that our approach outperformed existing methods in terms of accuracy, running time and scalability.

Metric Data Analysis Enhanced through Temporal Visualization

The human vision can naturally interpret data in spaces of 2 or 3 dimensions. When data is in higher dimensional spaces, in most cases the visualization is not intuitive. Regarding metric spaces, the interpretation is even harder, since they often do not have a direct spatial representation. However, the need to analyze how metric-represented data evolve over time is pretty common when one needs to understand several phenomena and in decision making processes, as it occurs in medical and agrometeorological applications. This paper presents three interactive techniques to visualize metric data that vary over time. Each one focus on a different way to interpret the temporal information. The first technique shows data evolving in a timeline axis. The second overlaps evolving snapshots of the space showing how the space varies regarding time. The last one does not treat temporal data as a dimension, it is used instead to define the similarity among complex data, employing the new concept of metric-temporal spaces, which seamlessly integrate time and metric data into a single similarity space. Visualization examples with real datasets are presented to show the usefulness of the proposed techniques.