Query Object Research Articles

Camera identification of multi-format devices

Photo Response Non-Uniformity (PRNU) based source camera attribution is an effective method to determine an image or a video’s origin camera. However, modern devices, especially smartphones, capture images and videos at different resolutions using the same sensor array, PRNU attribution can become ineffective as the camera fingerprint and query object can be misaligned. While capturing visual objects (either image or video), cameras may use different in-camera operations as well as they may use different parts of the sensor. In this paper, we investigate the problem of source camera attribution of a visual object by doing a thorough investigation of a comprehensive dataset, NYUAD Mixed Media Dataset. This investigation takes many factors into accounts, such as the fact that visual objects may have been captured using different resolution and aspect ratios. Furthermore, the visual objects may use different regions of the sensor, including the usage of boundary pixels for videos. Taking these various cases into account, we propose an efficient search which not only gives the state-of-the-art results but also performs significantly faster compared to existing methods.

High Intrinsic Dimensionality Facilitates Adversarial Attack: Theoretical Evidence

Machine learning systems are vulnerable to adversarial attack. By applying to the input object a small, carefully-designed perturbation, a classifier can be tricked into making an incorrect prediction. This phenomenon has drawn wide interest, with many attempts made to explain it. However, a complete understanding is yet to emerge. In this paper we adopt a slightly different perspective, still relevant to classification. We consider retrieval, where the output is a set of objects most similar to a user-supplied query object, corresponding to the set of $k$ -nearest neighbors. We investigate the effect of adversarial perturbation on the ranking of objects with respect to a query. Through theoretical analysis, supported by experiments, we demonstrate that as the intrinsic dimensionality of the data domain rises, the amount of perturbation required to subvert neighborhood rankings diminishes, and the vulnerability to adversarial attack rises. We examine two modes of perturbation of the query: either ‘closer’ to the target point, or ‘farther’ from it. We also consider two perspectives: ‘query-centric’, examining the effect of perturbation on the query’s own neighborhood ranking, and ‘target-centric’, considering the ranking of the query point in the target’s neighborhood set. All four cases correspond to practical scenarios involving classification and retrieval.

Processing Location-Based Aggregate Queries in Road Networks

In recent years, the research community has introduced various methods for processing the location-based queries on a single type of objects in road networks. However, in real-life applications user may be interested in obtaining information about different types of objects, in terms of their neighboring relationship. The sets of different types of objects closer to each other are termed the heterogeneous neighboring object sets (or HNO sets for short). To provide users with object information by considering both the spatial closeness of objects to the query object and the neighboring relationship between objects, we present useful and important location-based aggregate queries on finding the HNO sets in road net-works. The location-based aggregate queries are the shortest average distance query (SAvgDQ), the shortest minimal distance query (SMinDQ), the shortest maximal distance query (SMaxDQ), and the shortest sum distance query (SSumDQ). We first utilize a grid index to manage information of data objects and road networks, and then propose the SAvgDQ, SMinDQ, SMaxDQ, and SSumDQ processing algorithms, which are combined with the grid index to efficiently process the four types of location-based aggregate queries, respectively. A comprehensive set of experiments is conducted to demonstrate the efficiency of the proposed processing algorithms using real road network datasets.

PRESS: A personalised approach for mining top-k groups of objects with subspace similarity

Personalised analytics is a powerful technology that can be used to improve the career, lifestyle, and health of individuals by providing them with an in-depth analysis of their characteristics as compared to other people. Existing research has often focused on mining general patterns or clusters, but without the facility for customisation to an individual’s needs. It is challenging to adapt such approaches to the personalised case, due to the high computational overhead they require for discovering patterns that are good across an entire dataset, rather than with respect to an individual. In this paper, we tackle the challenge of personalised pattern mining and propose a query-driven approach to mine objects with subspace similarity. Given a query object in a categorical dataset, our proposed algorithm, PRESS (Personalised Subspace Similarity), determines the top-k groups of objects, where each group has high similarity to the query for some particular subspace. We evaluate the efficiency and effectiveness of our approach on both synthetic and real datasets.

A Parallelizing Method for Generation of Voronoi Diagram Using Contact Zone

Due to the recent popularization of the Geographic Information System (GIS), spatial network environments that can display the changes of spatial axes on mobile devices are receiving great attention. In spatial network environments, since a query object that seeks location information selects several candidate target objects based on the search conditions, we often use a k-nearest neighbor (kNN) search, which seeks several target objects near the query object. However, since a kNN search needs to find the kNN by calculating the distance from the query to all the objects, the computational complexity might become too large based on the number of objects. To reduce this computation time in a kNN search, many researchers have proposed a search method that divides regions using a Voronoi diagram. However, since conventional methods generate Voronoi diagrams for objects in order, the processing time for generating Voronoi diagrams might become too large when the number of objects is increased. In this paper, we propose a generation method of the Voronoi diagram by parallelizing the generation of Voronoi regions using a contact zone. Our proposed method can reduce the processing time of generating the Voronoi diagram by generating Voronoi regions in parallel based on the number of targets. Our evaluation confirmed that the processing time under the proposed method was reduced about 15.9\% more than conventional methods that are not parallelized.

Pruning techniques for parallel processing of reverse top-k queries

In this paper, we address the problem of processing reverse top-k queries in a parallel setting. Given a database of objects, a set of user preferences, and a query object q, the reverse top-k query returns the subset of user preferences for which the query object belongs to the top-k results. Although recently the reverse top-k query operator has been studied extensively, its CPU-intensive nature results in prohibitively expensive processing cost, when applied on vast-sized data sets. This limitation motivates us to explore a scalable parallel processing solution, in order to enable reverse top-k processing over distributed large sets of input data in reasonable execution time. We present an algorithmic framework for the problem, in which different algorithms can be instantiated, targeting a generic parallel setting. We describe a parallel algorithm (DiPaRT) that exploits basic pruning properties and is provably correct, as an instantiation of the framework. Furthermore, we introduce novel pruning properties for the problem, and propose DiPaRT+ as another instance of the algorithmic framework, which offers improved efficiency and scales gracefully. All algorithms are implemented in MapReduce, and we provide a wide set of experiments that demonstrate the improved efficiency of DiPaRT+ using data sets that are four orders of magnitude larger than those handled by centralized approaches.

A sketch-aided retrieval approach for incomplete 3D objects

With the growing amount of digital collections of visual CH data being available across different repositories, it becomes increasingly important to provide archaeologists with means to find relations and cross-correspondences between different digital records. In principle, existing shape- and image-based similarity search methods can aid such domain analysis tasks. However, in practice, visual object data are given in different modalities, and often only in incomplete or fragmented state, posing a particular challenge for conventional similarity search approaches. In this paper we introduce a methodology and system for cross-modal visual search in CH object data that addresses these challenges. Specifically, we propose a new query modality based on 3D views enhanced by user sketches (3D+sketch). This allows for adding new context to the search, which is useful e.g., for searching based on incomplete query objects, or for testing hypotheses on existence of certain shapes in a collection. We present an appropriately designed workflow for constructing query views from incomplete 3D objects enhanced by a user sketch, based on shape completion and texture inpainting. Visual cues additionally help users compare retrieved objects with the query. The proposed approach extends on a previously presented retrieval system by introducing improved retrieval methods, an extended evaluation including retrieval in a larger and richer data collection, and enhanced interactive search weight specification. We demonstrate the feasibility and potential of our approach to support analysis of domain experts in Archaeology and the field of CH in general.

Multi-objective spatial keyword query with semantics: a distance-owner based approach

Multi-objective spatial keyword query aims to find a set of objects that are reasonably distributed in spatial, with all query objectives to be satisfied. However, existing approaches mainly take the coverage of query keywords into account, while leaving the semantics of the textual data to be largely ignored. This limits us to return those rational results that are synonyms but morphologically different. To address this problem, this paper studies the problem of multi-objective spatial keyword query with semantics, and targets to return the object set that is optimum regarding to both spatial proximity and semantic relevance. Specifically, we take advantage of the probabilistic topic model and locality sensitive hashing (LSH), so that all query objectives can be satisfied in terms of their semantics. Afterwards, a novel indexing structure called LIR-tree is designed to integrate the spatial and semantic information of all objects in a balanced way. On top of the LIR-tree, we further propose a distance-owner based query processing algorithm, which provides tight bounds to achieve superb pruning effect in the searching phase. To speed up the processing, a distance owners based replacement strategy can be used to conduct approximate querying more efficiently. Empirical study based on a real dataset demonstrates the good effectiveness and efficiency of our proposed algorithms.

MaxiZone: Maximizing Influence Zone Over Geo-Textual Data

Given a geo-textual dataset $\mathbb {O}$ O , a set $\varphi$ φ of keywords, a query object $q \in \mathbb {O}$ q ∈ O , and an integer $k$ k , a reverse top- $k$ k keyword-based location query returns the influence zone $\mathcal {R}$ R of $q$ q such that $q$ q belongs to the result of a top- $k$ k spatial keyword query with query keywords $\varphi$ φ and any location in $\mathcal {R}$ R as arguments. For a query object $q$ q , the influence zone of $q$ q varies for different keywords $\varphi$ φ . Users may be interested in identifying the maximum influence zone of the query object. To this end, in this paper, we study the problem called MaxiZone that finds the keyword set maximizing the influence zone of a specified query object. The MaxiZone problem has many real-life applications, e.g., a business owner would like to identify the maximum influence zone so as to attract as many customers as possible. A straightforward way to tackle the MaxiZone problem is to compute the influence zone for every candidate keyword set. Obviously, this is infeasible if there are a large number of candidate keyword sets. We propose a more efficient index-centric algorithm together with a series of optimizations as well as a sampling-based algorithm, to facilitate the query processing. Moreover, we extend the proposed algorithms to address a variant of MaxiZone problem called $\tau$ τ -MaxiZone problem, which finds top- $\tau$ τ keyword sets having the maximum influence zones. Extensive empirical study using real-world datasets demonstrates the effectiveness and efficiency of our proposed algorithms.

HarmonyMoves: A Unified Prediction Approach for Moving Object Future Path

Trajectory prediction plays a critical role on many location-based services such as proximity-based marketing, routing services, and traffic management. The vast majority of existing trajectory prediction techniques utilize the object’s motion history to predict the future path(s). In addition to, their assumptions that the objects’ moving with recognized patterns or know their routes. However, these techniques fail when the history is unavailable. Also, these techniques fail to predict the path when the query moving objects lost their ways or moving with abnormal patterns. This paper introduces a system named HarmonyMoves to predict the future paths of moving objects on road networks without relying on their past trajectories. The system checks the harmony between the query object and other moving objects, after that if the harmony exists, this means that there are other objects in space moving like the query object. Then, a Markov Model is adopted to analyze this set of similar motion patterns and generate the next potential road segments of the object with their probabilities. If the harmony does not exist, HarmonyMoves considers this query object as abnormal object (object lost the way and needs support to return back known routes), for this purpose HarmonyMoves employed a new module to handle this case. A fundamental aspect of HarmonyMoves lies in achieving a high accurate prediction while performing efficiently to return query answers.

$DeepMotions$ : A Deep Learning System for Path Prediction Using Similar Motions

Trajectory prediction techniques play a serious role in many location-based services such as mobile advertising, carpooling, taxi services, traffic management, and routing services. These techniques rely on the object's motion history to predict the future path(s). As a consequence, these techniques fail when history is unavailable. The unavailability of history might occur for several reasons such as; history might be inaccessible, a recently registered user with no preceding history, or previously logged data is preserved for confidentiality and privacy. This paper presents a Bi-directional recurrent deep-learning based prediction system, named DeepMotions, to predict the future path of a query object without any prior knowledge of the object historical motions. The main idea of DeepMotions is to observe the moving objects in the vicinity that have similar motion patterns of the query object. Then use those similar objects to train and predict the query object's future steps. To compute similarity, we propose a similarity function that is based on the KNN algorithm. Extensive experiments conducted on real data sets confirm the efficient performance and the quality of prediction in DeepMotions with up to 96% accuracy.

Efficient processing of neighboring skyline queries with consideration of distance, quality, and cost

Currently, many of the processing techniques for the location-based queries provide information of a single type of spatial objects, based on their spatial closeness to the query object. However, in real-life applications the user may be interested in obtaining information about different types of objects, in terms of their quality, cost, and neighboring relationship. We term the different types of objects with better quality and closer to each other the Neighboring skyline set (or NS set). Three new types of location-based queries, the Distance-based neighboring skyline query (Dist-NS query), the Cost-based neighboring skyline query (Cost-NS query), and the Budget-based neighboring skyline query (BGT-NS query), are presented to determine the NS sets according to user’s specific requirement. A R-tree-based index, the $$R^{a,c}$$-tree, is first designed to manage each type of objects with their locations, attributes, and costs. Then, a simultaneous traversal of the $$R^{a,c}$$-trees built on different types of objects is employed with several pruning criteria to prune the non-qualifying object sets as early as possible, so as to improve the query performance. Extensive experiments using the synthetic dataset demonstrate the efficiency and the effectiveness of the proposed algorithms.

Mobile P2P-Based Skyline Query Processing over Delay-Tolerant Networks

Skyline query-processing techniques considering various properties in peer to peer (P2P)-based services have become a recent topic of research. In this paper, we propose a new skyline query-processing scheme to improve the query-processing performance and accuracy in a mobile P2P service over delay-tolerant networks. The proposed scheme collects data on the query object from neighboring nodes and establishes a local skyline through static properties to reduce query-processing costs. To improve the query accuracy in a non-uniform distribution environment, the query-dissemination range is expanded by enforcing a query-dissemination range expansion. The performance evaluation conducted to verify the superiority of the proposed scheme demonstrates that it has a better performance compared to the existing schemes.

Efficient processing of moving collective spatial keyword queries

As a major type of continuous spatial queries, the moving spatial keyword queries have been studied extensively. Most existing studies focus on retrieving single objects, each of which is close to the query object and relevant to the query keywords. Nevertheless, a single object may not satisfy all the needs of a user, e.g., a user who is driving may want to withdraw money, wash her car, and buy some medicine, which could only be satisfied by multiple objects. We thereby formulate a new type of queries named the moving collective spatial keyword query (MCSKQ). This type of queries continuously reports a set of objects that collectively cover the query keywords as the query moves. Meanwhile, the returned objects must also be close to the query object and close to each other. Computing the exact result set is an NP-hard problem. To reduce the query processing costs, we propose algorithms, based on safe region techniques, to maintain the exact result set while the query object is moving. We further propose two approximate algorithms to obtain even higher query efficiency with precision bounds. All the proposed algorithms are also applicable to MCSKQ with weighted objects and MCSKQ in the domain of road networks. We verify the effectiveness and efficiency of the proposed algorithms both theoretically and empirically, and the results confirm the superiority of the proposed algorithms over the baseline algorithms.

Learning Style Compatibility Between Objects in a Real‐World 3D Asset Database

AbstractLarge 3D asset databases are critical for designing virtual worlds, and using them effectively requires techniques for efficient querying and navigation. One important form of query is search by style compatibility: given a query object, find others that would be visually compatible if used in the same scene. In this paper, we present a scalable, learning‐based approach for solving this problem which is designed for use with real‐world 3D asset databases; we conduct experiments on 121 3D asset packages containing around 4000 3D objects from the Unity Asset Store. By leveraging the structure of the object packages, we introduce a technique to synthesize training labels for metric learning that work as well as human labels. These labels can grow exponentially with the number of objects, allowing our approach to scale to large real‐world 3D asset databases without the need for expensive human training labels. We use these synthetic training labels in a metric learning model that analyzes the in‐engine rendered appearance of an object—combining geometry, material, and texture—whereas prior work considers only object geometry, or disjoint geometry and texture features. Through an ablation experiment, we find that using this representation yields better results than using renders which lack texture, materiality, or both.

Direction-Aware Continuous Moving K-Nearest-Neighbor Query in Road Networks

Continuous K-nearest neighbor (CKNN) queries on moving objects retrieve the K-nearest neighbors of all points along a query trajectory. They mainly deal with the moving objects that are nearest to the moving user within a specified period of time. The existing methods of CKNN queries often recommend K objects to users based on distance, but they do not consider the moving directions of objects in a road network. Although a few CKNN query methods consider the movement directions of moving objects in Euclidean space, no efficient direction determination algorithm has been applied to CKNN queries over data streams in spatial road networks until now. In order to find the top K-nearest objects move towards the query object within a period of time, this paper presents a novel algorithm of direction-aware continuous moving K-nearest neighbor (DACKNN) queries in road networks. In this method, the objects’ azimuth information is adopted to determine the moving direction, ensuring the moving objects in the result set towards the query object. In addition, we evaluate the DACKNN query algorithm via comprehensive tests on the Los Angeles network TIGER/LINE data and compare DACKNN with other existing algorithms. The comparative test results demonstrate that our algorithm can perform the direction-aware CKNN query accurately and efficiently.

Distributed Processing of Location-Based Aggregate Queries Using MapReduce

The location-based aggregate queries, consisting of the shortest average distance query (SAvgDQ), the shortest minimal distance query (SMinDQ), the shortest maximal distance query (SMaxDQ), and the shortest sum distance query (SSumDQ) are new types of location-based queries. Such queries can be used to provide the user with useful object information by considering both the spatial closeness of objects to the query object and the neighboring relationship between objects. Due to a large amount of location-based aggregate queries that need to be evaluated concurrently, the centralized processing system would suffer a heavy query load, leading eventually to poor performance. As a result, in this paper, we focus on developing the distributed processing technique to answer multiple location-based aggregate queries, based on the MapReduce platform. We first design a grid structure to manage information of objects by taking into account the storage balance, and then develop a distributed processing algorithm, namely the MapReduce-based aggregate query algorithm (MRAggQ algorithm), to efficiently process the location-based aggregate queries in a distributed manner. Extensive experiments using synthetic and real datasets are conducted to demonstrate the scalability and the efficiency of the proposed processing algorithm.

Tree-based mining contrast subspace

All existing mining contrast subspace methods employ density-based likelihood contrast scoring function to measure the likelihood of a query object to a target class against other class in a subspace. However, the density tends to decrease when the dimensionality of subspaces increases causes its bounds to identify inaccurate contrast subspaces for the given query object. This paper proposes a novel contrast subspace mining method that employs tree-based likelihood contrast scoring function which is not affected by the dimensionality of subspaces. The tree-based scoring measure recursively binary partitions the subspace space in the way that objects belong to the target class are grouped together and separated from objects belonging to other class. In contrast subspace, the query object should be in a group having a higher number of objects of the target class than other class. It incorporates the feature selection approach to find a subset of one-dimensional subspaces with high likelihood contrast score with respect to the query object. Therefore, the contrast subspaces are then searched through the selected subset of one-dimensional subspaces. An experiment is conducted to evaluate the effectiveness of the tree-based method in terms of classification accuracy. The experiment results show that the proposed method has higher classification accuracy and outperform the existing method on several real-world data sets.

Direction-based spatial skyline for retrieving surrounding objects

Retrieval of surrounding data objects has many potential applications in spatial databases including nearby point-of-interest retrieval surrounding a user and digital gaming. This paper presents a novel query, called direction-based spatial skyline (DSS), for retrieving surrounding objects in spatial databases. A DSS query retrieves all surrounding objects from a spatial database that are not directionally dominated by other data objects w.r.t. a given query object. The proposed DSS query is rotationally invariant, fair and stable. The paper also presents efficient algorithms for processing DSS queries in spatial databases by designing novel data pruning techniques using R-Tree data indexing scheme. We conduct extensive experiments using both real and synthetic datasets to demonstrate the effectiveness and efficiency of the proposed algorithms.

A Training-Free Approach for Generic Object Detection

We present an approach for generic object detection using single query image for finding and locating visually similar objects from target images. The key challenge here is describing an object class using only one query without any training. Our approach is based on computation of Local Self-Similarity descriptors which captures local internal geometric layout within an image and is good representative of object class. We propose to use only predefined landmark points from query image which significantly improves performance of detection. We also present few novel ideas for selection of informative descriptors from the set of all descriptors of the test image to reduce computational expense in feature matching. The algorithm yields Hough-style similarity surface indicating likelihood of presence of the query object at every location. Presence and location of objects are finalized by employing two significance tests followed by non-maxima suppression. We evaluate results of the proposed approach on UIUC car dataset and achieve higher accuracy in comparison with earlier training-free approaches. Results are also analysed on ETHZ shape classes dataset which accounts for large intra-class variations, scale variations, and clutter. Performance of detection on this challenging dataset having diverse contexts demonstrates robustness and efficacy of the algorithm.