Archive for the ‘publications’ Category
Posted by eduardovalle on Wednesday, April 21, 2010
Our paper, “MONORAIL: A Disk-Friendly Index for Huge Descriptor Databases” was accepted at the upcoming IAPR Internation Conference on Pattern Recognition — ICPR 2010. Here is the abstract:
We propose MONORAIL, an indexing scheme for very large multimedia descriptor databases. Our index is based on the Hilbert curve, which is able to map the high-dimensional space of those descriptors to a single dimension. Instead of using several curves to mitigate boundary effects, we use a single curve with several surrogate points for each descriptor. Thus, we are able to reduce the random accesses to the bare minimum. In a rigorous empirical comparison with another method based on multiple surrogates, ours shows a significant improvement, due to our careful choice of the surrogate points.
I am particularly proud of this paper, not only because of the method itself, but also because of the experimental design we propose for the validation. I have been studying for more than a year the topics of Design of Experiments, statistical tests and validation. This is the first of a crop of publications that are employing those rigorous evaluation tools, which, though commonplace in other fields, are still seldom used in Computer Sciences.
Posted in publications, science | Tagged: conference, design of experiments, Fernando Akune, kNN search, monorail, paper, publication, Ricardo Torres | Leave a Comment »
Posted by eduardovalle on Friday, January 22, 2010
Our paper, “Learning to Rank for Content-Based Image Retrieval” , was accepted at the upcoming ACM Multimedia Information Retrieval Conference (MIR 2010). The first author is the M.Sc. student Fábio Faria, and the paper was co-authored with my Post Doc supervisor Ricardo Torres and several of our partners from UFMG, including Marcos Gonçalves, with whom we have an ongoing cooperation.
Here is the abstract:
“In Content-based Image Retrieval (CBIR), accurately ranking the returned images is of paramount importance, since users consider mostly the topmost results. The typical ranking strategy used by many CBIR systems is to employ image content descriptors, so that returned images that are most similar to the query image are placed higher in the rank. While this strategy is well accepted and widely used, improved results may be obtained by combining multiple image descriptors. In this paper we explore this idea, and introduce algorithms that learn to combine information coming from different descriptors. The proposed learning to rank algorithms are based on three diverse learning techniques: Support Vector Machines (CBIR-SVM), Genetic Programming (CBIR-GP), and Association Rules (CBIR-AR). Eighteen image content descriptors (color, texture, and shape information) are used as input and provided as training to the learning algorithms. We performed a systematic evaluation involving two complex and heterogeneous image databases (Corel e Caltech) and two evaluation measures (Precision and MAP). The empirical results show that all learning algorithms provide significant gains when compared to the typical ranking strategy in which descriptors are used in isolation. We concluded that, in general, CBIR-AR and CBIR-GP outperforms CBIR-SVM. A fine-grained analysis revealed the lack of correlation between the results provided by CBIR-AR and the results provided by the other two algorithms, which indicates the opportunity of an advantageous hybrid approach.”
I will be travelling to Philadelphia on late March to present the poster. I am very excited about this upcoming trip to the United States, where I am to meet several friends and colleagues, but at the same time, worried about the radicalization of air security rules and the exaggeration of perception of threats. Have we got so scared to die that we decided instead not to live ?
Posted in publications, science | Tagged: paper, conference, publication, cbir, DCC / UFMG, Ricardo Torres, poster, MIR, learn to rank, machine learning, Fábio Faria, Marcos Gonçalves | Leave a Comment »
Posted by eduardovalle on Wednesday, November 11, 2009
- I am visiting the Mathematical and Computing Sciences Instiute of the University of São Paulo (ICMC – USP) at São Carlos, on Tuesday, November 24. I’ll visit Professors Agma Traina and Caetano Traina Jr. and their team — they have a long experience on similarity search, and have proposed the Slim-Tree, a metric space technique. I’ll give a talk on the Scalability Issues of Multimedia Information Retrieval.
- That same week, I am visiting the Mathematics and Statistics Institute of the University of São Paulo (IME – USP) at São Paulo, on Friday, November 27. I’ll visit Professor Roberto Cesar Jr. and his student, the Ph.D. candidate Marcelo Hashimoto, both of whom I had the pleasure to meet on last SIBGRAPI. I’ll also give a talk there, the title will be announced soon. (EDIT 26/11: The talk will be about “Similarity search for Multimedia” and will be heavily based on my SBBD 2009 Tutorial — more details can be found on my Google Calendar).
- The slides of my (and Prof. Matthieu Cord’s) Tutorial on Similarity Search and High-Dimensional Indexing, which I gave at this year’s SBBD, are finally available.
- I am still working to make available the slides of my (and Prof. Matthieu Cord’s) Tutorial on Advanced Techniques for CBIR, which I gave at this year’s SIBGRAPI. The companion article to the tutorial, however, is already available.
Posted in publications | Tagged: kNN search, publication, cbir, seminar, sibgrapi, tutorial, SBBD, icmc-usp, ime-usp, similarity search | Leave a Comment »
Posted by eduardovalle on Sunday, September 20, 2009
I guess that for all people involved, DocEng’09 was a success. Like last year, the conference was small — I think that we were 60 or 70 participants — but the quality of the works presented was high, and the scientific exchange was extremely interesting. In DocEng, you get to meet everyone individually, something which is unfeasible at large-scale conferences.
Thematically, the conference has a broad scope, centered around the representation, processing, analysis, storage and retrieval of documents. My main research topic concerns the retrieval of multimedia documents, and is somewhat at the fringe of the conference theme. Nevertheless, people seemed genuinely interested and I’ve got many useful insights and suggestions.
* * *
I have just arrived at Paris, where I will meet my former Ph.D. supervisor Prof. Matthieu Cord, among other colleagues. I intend to advance our research on high-dimensional multimedia indexing and large scale multimedia retrieval. I am also giving a talk about my current research pursuits at the ETIS labs, on Cergy-Pontoise, next Tuesday, September 22nd.
If you use Google Calendar you can save the date by clicking below:
Posted in publications, science | Tagged: cooperation, doceng, etis, France, Germany, lip6, publication, seminar | Leave a Comment »
Posted by eduardovalle on Wednesday, September 9, 2009
My tutorial Similarity Search and Indexing for High-Dimensional Data has been accepted on SBBD 2009 (The Brazilian Symposium on Databases). Here’s the abstract:
Searching by similarity is a critical operation on many systems, and thus has attracted the attention of many disciplines in Computer Sciences, including Computational Geometry, Machine Learning, Multimedia and, of course, Databases. To perform efficiently, similarity search requires the support of indexing, which suffers from the infamous “curse of the dimensionality”. In this tutorial we will introduce the challenges of indexing and searching high-dimensional data, and present the most recent tools available to “tame the curse”. At the end, the audience will have a good grasp of the current state of the art, the most promising research trends and the challenges still faced by the technology.
The tutorials, as I understand, are open to all participants on the conference. Mine will be held on Wednesday, October 7th from 14h40 to 18h20, with a 20′ coffee-break. If you use Google calendar, you can save the date by clicking on the button below.
* * *
I’ve unintentionally let an awful lot of of time pass since my last post — the move to Campinas (and to UNICAMP) has been wonderful, but also laborious. I thought that after moving across countries three times, moving across states would be a piece of cake, but it seems that, no matter the distance, moving is always a lot of hassle!
EDIT 11/11/09: The tutorial presentation, for the moment without narrative, is available on my talks and courses page.
Posted in career, publications | Tagged: indexing, kNN search, publication, SBBD, tutorial, UNICAMP | Leave a Comment »
Posted by eduardovalle on Sunday, July 5, 2009
I had two short papers accepted on DocEng 2009.
One, co-authored with my French partners Dr. David Picard and Prof. Matthieu Cord, is about the difficult problem of enforcing geometric consistency in vote-counting based CBIR when there are too many outliers — a situation we encounter routinely in our iTowns project. Here’s the title and abstract:
Geometric Consistency Checking for Local-Descriptor Based Document Retrieval — In this paper, we evaluate different geometric consistency schemes, which can be used in tandem with an efficient architecture, based on voting and local descriptors, to retrieve multimedia documents. In many contexts the geometric consistency enforcement is essential to boost the retrieval performance. Our empirical results show however, that geometric consistency alone is unable to guarantee high-quality results in databases that contain too many non-discriminating descriptors.
The other, co-authored with my Brazilian colleagues Flávio Bertholdo and Prof. Arnaldo Araújo, proposes a new method for contrast enhancement in degraded historical documents, which takes into account the structure of the the document:
Layout-Aware Limiarization for Readability Enhancement of Degraded Historical Documents — In this paper we propose a technique of limiarization (also known as thresholding or binarization) tailored to improve the readability of degraded historical documents. Limiarization is a simple image processing technique, which is employed in many complex tasks like image compression, object segmentation and character recognition. The technique also finds applications on itself: since it results in a high-contrast image, in which the foreground is clearly separated from the background, it can greatly improve the readability of a document, provided that other attributes (like character shape) do not suffer. Our technique exploits statistical characteristics of textual documents and applies both global and local thresholding. Under visual inspection on experiments made in a collection of severely degraded historical documents, it compares favorably with the state of the art.
DocEng 2009 will be held in Munich, Germany on September 15–18.
EDIT 23/07: The preprints are now available in my publications page.
Posted in publications, science | Tagged: paper, doceng, publication, cultural heritage, cbir, limiarization, iTowns, geometric consistency | Leave a Comment »
Posted by eduardovalle on Wednesday, June 24, 2009
My tutorial proposal “Advanced Techniques in CBIR: Local Descriptors, Visual Dictionaries and Bags of Features” has been accepted on the Brazilian Symposium on Computer Graphics and Image Processing (SIBGRAPI 2009), which will be held at the Pontifícia Universidade Católica do Rio de Janeiro (PUC-Rio), Brazil between October 11th and 14th, 2009. The tutorial itself will probably happen on October 11th.
Here’s the abstract:
“Local descriptors have been extensively used in CBIR systems, where their robustness to intense geometric and photometric transformations allows the identification of a target object/image with great reliability. However, due to their excessive discriminating power, their application to the retrieval of complex categories is challenging. The introduction of the technique of ‘visual dictionaries’ (also known as ‘dictionary of visual terms’) is an important step towards the conciliation between the robustness of local descriptors and the flexibility of generalization needed by complex queries. As a bonus, we become able to employ advanced retrieval techniques which were so far available only for textual data.
The importance of local descriptors in the current practice of CBIR can hardly be overestimated, and the introduction of the ‘visual dictionary’ technique has allowed to employ them on complex category search, a context where they were previously considered too discriminating to work effectively. In this tutorial we will address all the conceptual components of a powerful framework — the CBIR system architecture itself, the local descriptors and the techniques which allow their advanced application. At the end, the audience will have a good idea of the current state of art, the most promising research trends and the challenges still faced by the technology.”
The tutorial will take 3 hours and will be given either in English or Portuguese, according to the composition of the classroom.
EDIT 11/11/2009: The tutorial survey is available on my publications page.
Posted in publications | Tagged: publication, cbir, sibgrapi, tutorial | Leave a Comment »
Posted by eduardovalle on Tuesday, October 28, 2008
I’ve presented the paper “High-Dimensional Descriptor Indexing for Large Multimedia Databases” in the Session 5C : Information Retrieval — Multilingual and Multimedia of the 17th ACM Conference on Information and Knowledge Management — CIKM 2008. This paper presents the third (and most interesting, in my opinion) method for multidimentional indexing proposed in my Ph.D. thesis.
Here’s the abstract:
“In this paper we address the subject of large multimedia database indexing for content-based retrieval. We introduce multicurves, a new scheme for indexing high-dimensional descriptors. This technique, based on the simultaneous use of moderate-dimensional space-filling curves, has as main advantages the ability to handle high-dimensional data (100 dimensions and over), to allow the easy maintenance of the indexes (inclusion and deletion of data), and to adapt well to secondary storage, thus providing scalability to huge databases (millions, or even thousands of millions of descriptors). We use multicurves to perform the approximate k nearest neighbors search with a very good compromise between precision and speed. The evaluation of multicurves, carried out on large databases, demonstrates that the strategy compares well to other up-to-date k nearest neighbor search strategies. We also test multicurves on the real-world application of image identification for cultural institutions. In this application, which requires the fast search of a large amount of local descriptors, multicurves allows a dramatic speed-up in comparison to the brute-force strategy of sequential search, without any noticeable precision loss.”
The paper and some related material can be found in my publications page.
Posted in publications, science | Tagged: paper, cikm, multicurves, kNN search, conference, image identification, retroblogged, thesis, publication | Leave a Comment »
Posted by eduardovalle on Thursday, September 18, 2008
I’ve presented the paper “Fast Identification of Visual Documents Using Local Descriptors” on the 8th ACM Symposium on Document Engineering — DocEng 2008. This paper had a stronger emphasis in the application (image identification, also known as copy detection or — my least favourite nomenclature — near-duplicate detection) than in the technicalities of multimedia indexing and descriptor matching (which were, in the end of the day, the staple of my Ph.D. work). But it was the opportunity to tell the world about the projection KD-forest, the second (among three) original technique for high-dimensional indexing.
Anyway, here’s the abstract: “In this paper we introduce a system for the identification of visual documents. Since it stems from content-based document indexing and retrieval, our system does not need to rely on textual annotations, watermarks or other metadata, which can be missing or incorrect. Our retrieval system is based on local descriptors, which have been shown to provide accurate and robust description. Because of the high computational costs associated to the matching of local descriptors, we propose Projection KD-Forest: an indexing technique which allows efficient approximate k nearest neighbors search. Experiments demonstrate that the Projection KD-Forest allows the system to provide prompt results with negligible loss on accuracy. The Projection KD-Forest also compares well when contrasted to other strategies of k nearest neighbors search.”
I guess that the dream of all conference organisers is to amass hundreds of participants and to have an acceptance rate as close to zero as possible. This edition of DocEng was nothing like that — I think that we were 60 or 70 counting everyone, authors and participants. Yet, in terms of exchange of scientific ideas, it was one of the best conferences I’ve ever participated.
The paper, and related material, can be found in my publications page.
Posted in publications, science | Tagged: conference, doceng, image identification, kNN search, paper, publication, retroblogged, thesis | Leave a Comment »
Posted by eduardovalle on Thursday, June 12, 2008
I’ve passed the viva-voce defense of my Ph.D. thesis “Local-Descriptor Matching for Image Identification Systems”, thus completing the program.
I won’t pretend that it is not a relief to have survived the process, it certainly wasn’t easy!
But I have been lucky enough to work with a subject about which I am passionate, and with people who are amazing. So, even though I’ve got the much sought after title of “Dr.” I expect to keep on the same fruitful research track (subject, team) for a while.
The abstract of the thesis: “Image identification (or copy detection) consists in retrieving the original from which a query image possibly derives, as well as any related metadata, such as titles, authors, copyright information, etc. The task is challenging because of the variety of transformations that the original image may have suffered. Image identification systems based on local descriptors have shown excellent efficacy, but often suffer from efficiency issues, since hundreds, even thousands of descriptors, have to be matched in order to find a single image. The objective of our work is to provide fast methods for descriptor matching, by creating efficient ways to perform the k-nearest neighbours search in high-dimensional spaces. In this way, we can gain the advantages from the use of local descriptors, while minimising the efficiency issues. We propose three new methods for the k-nearest neighbours search: the 3-way trees — an improvement over the KD-trees using redundant, overlapping nodes; the projection KD-forests — a technique which uses multiple moderate dimensional KD-trees; and the multicurves, which is based on multiple moderate dimensional Hilbert space-filling curves. Those techniques try to reduce the amount of random access to the data, in order to be well adapted to the implementation in secondary memory.”
The full text, and other goodies, are available in my publications page.
Posted in career, publications, science | Tagged: defense, image identification, kNN search, publication, thesis, Université de Cergy-Pontoise | Leave a Comment »
