Website last updated: July 15th, 2014.

Project - About ETFImageSearch

ETFImageSearch a.k.a ETFIS is an open-source platform for experimenting with Content-Based Image Retrieval (CBIR) algorithms. It enables to index images in a folder, search them by example image, and also run some benchmarks, all using a nice GUI. It is not intended as an end-user product but if you find it useful you can use it that way too. ETFIS also provides a CLI interface which enables scripting or invoking from a web-service.

ETFIS is implemented in C++ and uses Qt for rendering user interface, though it can be compiled "headless". ETFIS includes a "hacked" version of libjpeg which enables to implement algorithms in compressed domain that deliver superior performance to other methods. Our goal is to implement many different retrieval algorithms described in literature. Also ETFIS experimentally supports parallel processing, although this code is not currently in the download version. Experiments are successfully performed in parallel on a cluster of 25 computers.

Project is lead by Vedran Ljubović, a PhD student. His menthor is Haris Šupić.

ETFIS is licensed under GNU General Public License v3.


Screenshot of current version as of September 30th 2013. Click to enlarge

Experimental results

So which image retrieval method is the best? We hope to give an answer to this question one day. Stay tuned for continued experiments!

Experiment 3 (Last update: March 2014): Optimal parameters for tuning of fuzzy color histograms in a hybrid color model
* Project files on Google Docs
Related paper: "Hybrid Color Model for Image Retrieval Based on Fuzzy Histograms"

Experiment 2 (Last update: August 2013): Determine the optimal parameters for fuzzy color matching in 10-bin HSL histogram.
* Excel format * OpenDocument format
Related paper: "A Compact Feature for Image Retrieval in HSL Model"

Experiment 1 (Last update: August 2nd, 2013.): Determine the optimal parameters for construction of color histogram.
* Excel format (1.1 MB) * OpenDocument format (237.7 kB)
Related paper: "Comparative Study of Color Histograms as Global Feature for Image Retrieval"
Discussion: In general we got the best results with histogram in HSV color space with 256 bins (color 16*4*4 quantization) and using Matsushita distance.

 

Notes: Files given above contain results from a large number of experiments performed using ETFImageSearch and LIRE applications. Please consult our papers cited below for details on methodology. Results presented here differ somewhat from the ones given in our papers. The reasons are following:

Overall the differences in results do not affect the findings of published papers in a significant way. We invite members of scientific community to download ETFImageSearch and reproduce our results, or send bug fixes, refinements or methodology changes. Note that almost every paper uses slightly different retrieval methodology.

Recent papers related to project

Bibliography in BibTeX format

Legal note: All papers published on IEEE conferences are reproduced here under IEEE Copyright and Consent Form which states: "Personal Servers. Authors and/or their employers shall have the right to post the accepted version of IEEE-copyrighted articles on their own personal servers or the servers of their institutions or employers without permission from IEEE, provided that the posted version includes a prominently displayed IEEE copyright notice and, when published, a full citation to the original IEEE publication, including a link to the article abstract in IEEE Xplore. Authors shall not post the final, published versions of their papers."

Papers published on ACM conferences are reproduced according to ACM copyright transfer which states: "Each of the Employer/Author(s) retains the following rights: The right to post author-prepared versions of the Work covered by the ACM copyright in a personal collection on their own home page, on a publicly accessible server of their employer and in a repository legally mandated by the agency funding the research on which the Work is based. Such posting is limited to noncommercial access and personal use by others, and must include the following notice both embedded within the full text file and in the accompanying citation display as well: “© ACM, (YEAR). This is the author’s version of the work. It is posted here by permission of ACM for your personal use. Not for redistribution. The definitive version was published in PUBLICATION, {VOL#, ISS#, (DATE)} http://doi.acm.org/10.1145/{nnnnnn.nnnnnn}”. (You may find the nnnnnn.nnnnnn number for your article DOIs on its citation page in the ACM Digital Library.)"

Errata:

Downloads - ETFImageSearch

Binaries:

Coming soon :)

 

Source code:

ETFImageSearch-2014-03-11.tar.gz (417 kB)
Using Qt 4.8
Status: beta
 

ETFImageSearch on GitHub

Build requirements (Windows):
- We recommend to build ETFImageSearch using Qt SDK.
- Alternatively, use: Qt libraries 4.8.5/mingw and mingw. Qt Creator is optional.

Build requirements (Linux):
- Ubuntu: sudo apt-get install build-essential libqt4-dev qtcreator qt-dev-tools
- Similar approach for other distributions
- To build, enter ETFImageSearch directory and type: "qmake && make && make install"

Latest version of ETFIS is using libANN as an optional kNN indexer. You can install it (package named libann-dev or similar), or disable it by editing file ETFImageSearch.pro.

Related projects