Looking back at 20 years of ICPR conferences
As the 20th ICPR conference approaches fast, I hope that this article finds you in the midst of travel preparations for Istanbul. Pre-ICPR time is always loaded with anticipation, setting expectations, selecting which presentations to attend, and of course making sure that you will make the most out of your trip. This time, ICPR is even more special: our conference has reached an important milestone in its history via its 20th edition. To celebrate this, let’s have a look at how our conference has evolved during the last 20 years.
Our analysis covers the ICPR proceedings from 1988 to 2008. Although it would have been extremely interesting to look at all twenty conferences, we have been limited by the unavailability of web-based electronic records for years preceding 1988. An excellent historical overview on the early years of ICPR can be found in the Detailed History of the IAPR, written by Herbert Freeman, IAPR Fellow, and available on the IAPR web site.
Our analysis focuses on a series of characteristics like the number of published papers, the number of authors, patterns of collaboration, distribution of papers per author or community, etc. These characteristics are studied by the fields of Bibliometrics and Scientometrics.
Growth of the ICPR Conference
“The More, the Merrier”
We study the growth of the ICPR conference via the simple measure of accepted papers. Other measures that could be considered include the number of conference attendants, and the number of tutorials and satellite workshops; however, such measures cannot be retrieved from web-based information for the entire period of interest. As we can see from Table 1, the number of papers has been steadily increasing from one year to the next, with two exceptions. ICPR 2000 (Barcelona), and ICPR 2006 (Hong Kong) have both registered more accepted papers than their successors. The overall trend shows however an increasing pattern: since 1988, the number of accepted papers has almost tripled.
Table 1. Number of Accepted Papers for ICPR 1998-2008
The growth of the ICPR conference needs to be discussed in correlation with the selectivity of the conference. Indeed, a quantitative growth in the number of accepted papers would not mean much without maintaining or increasing the quality of these papers and implicitly the selectivity of the conference. The acceptance rate is a widely used measure for the selectivity of academic conferences; however, one needs to be aware of the skews introduced by the self-selection effect. Self-selection occurs when authors submit their work to more selective venues only if they think that they can have some chance of being accepted.
Table 2 summarizes data on the acceptance rates for 2000 onwards; this data was collected from the Welcome Messages of the conferences co-chairs [1-5]. Welcome messages for conferences prior to 2000 were not retrievable from web-based electronic resources. As one can see from Table 2, the selectivity of the ICPR conference can be considered stable, with fluctuations between 65% and 53% for the past four conferences. There is a wide gap between acceptance rates for oral presentations and poster presentations, which is consistent with the rates measured for other reputable venues in similar fields (CVPR, ICCV, ECCV).
Table 2. Acceptance Rate Statistics for ICPR 2000-2008
“Coming together is a beginning. Keeping together is progress. Working together is success.” Henry Ford
In order to discover collaboration patterns that characterize the ICPR community, data mining methods were used for performing authorship and co-authorship analysis. These data mining methods will be briefly discussed at the end of this article.
Authorship analysis involves statistics about the number of papers contributed to ICPR per author. This gives us an idea about the openness of the ICPR community to newcomers and new ideas. Via data mining, we have found that from a total number of 11,776 authors at ICPR1988-2008:
8,382 have one paper, or 71.18 % of the total number of papers
1,775 have two papers, or 15.07 % of the total number of papers
646 have three papers, or 5.49 % of the total number of papers
973 have four or more, or 8.26 % of the total number of papers
From these percentages, it is easy to see that the influx of newcomers is an important feature of ICPR.
We were also interested in performing authorship analysis at a national level, i.e. in determining for a given country the total number of papers that have at least one author from that country. Table 3 summarizes these results in the form of a Top 20 of the most productive countries.
Table 3. Papers contributed to ICPR per country from 1988 to 2008 – top 20
*One should note that Hong Kong’s contributions were counted separately from China and UK since Hong Kong’s transfer of sovereignty to China occurred on July 1 1997, roughly in the middle of the analyzed time interval (1988-2008). Therefore, an ISI ranking for Hong Kong is not available either.
As one may see from Table 3, USA ranks first with a significant advance from the runner-up. The next most productive countries are Japan, China, and France. This uneven distribution of papers per country is not unique to ICPR. According to the Essential Science Indicators of the ISI Web of Knowledge (In-cites.com, 2008) USA produces around 30% of the total number of academic publications worldwide, followed by Japan, Germany and UK with approximately 6-8% each. One may conclude that there is a good correlation between the ICPR ranking of the most productive countries and the ISI ranking.
Co-authorship analysis focuses on examining the collaborative relationships that are expressed as co-authored papers at ICPR. These relationships are examined at individual levels. We were interested in finding out the percentage of ICPR papers that are co-authored (2 authors, 3 authors, and more). Table 4 summarizes this study.
As one may see from Table 4, the dominant pattern of ICPR co-authorship consists in having 2 co-authors per paper (39.27% of the total number of papers in 1988-2008). This is followed by the patterns consisting in 3 co-authors (30.66%), 3 co-authors (13.53%), and single-authored papers (9.54%).
These collaborative patterns are not surprising. Our hypothesis is that numerous ICPR papers are produced as a result of collaboration between a graduate student and their supervisor, hence the most popular pattern consisting in 2 co-authors per paper. Papers involving 3 or more authors involve either more complex projects with more than one graduate student or interdisciplinary research. More research needs to be done to validate our hypothesis. It would also be interesting to look at how collaboration patterns evolve from one conference to the next, and whether these patterns vary by tracks or research topics within ICPR.
Scientific Impact of ICPR
“If I have seen further, it is by standing on the shoulder of giants.” Isaac Newton
The field of pattern recognition is an extremely active area of research with numerous links to other disciplines. Therefore, we are interested in determining how ICPR articles are influencing the evolution of this field. We have assembled the Top 10 of the most cited ICPR papers in Table 5. This data was gathered using Google Scholar.
Our methodology for data analysis
In order to complete our analysis of the ICPR conference, a comprehensive index of all papers needs to be created. The IEEE Xplore database contains all ICPR proceedings since 1988. It is then possible to create the necessary index list by parsing the html documents from IEEE Xplore. It has been determined that XML is the best solution for storing this index data because of its data layout structures and portability between programming languages.
An initial XML index structure was manually created as follows:
With the above structure it is possible to automatically query the IEEE Xplore site for indexing articles included in each proceedings. Information about authors, affiliation, countries, etc. is recorded as meta-data associated with the article. Errors occur when the affiliation data is not presented in a standard format. These errors are handled as exceptions. Technical information about error handling is available on demand. The index structure that we have generated allows for more advanced analysis of conference proceedings, such as retrieval and grouping of articles that cite common references, analysis of international collaborations in ICPR etc. This is future work, which will be submitted for publication in a peer-reviewed journal.
 "Message from the ICPR2000 Local Organizing Committee Co-Chairs," icpr, pp.xxiii, 15th International Conference on Pattern Recognition (ICPR'00), 2000
 "Message from the General Chairs," icpr, pp.xxii, 16th International Conference on Pattern Recognition (ICPR'02), 2002.
 "Message from the Chairs," icpr, pp.xix, 17th International Conference on Pattern Recognition (ICPR'04), 2004.
 “Message from the ICPR 2006 General Co-Chairs,” icpr, 18th International Conference on Pattern Recognition (ICPR'06), 2006.
 “Welcome Message”, icpr, 19th International Conference on Pattern Recognition (ICPR’08), 2008.
 “Most cited ICPR,” http://scholar.google.ca/scholar?
attern+recognition&as_ylo=&as_vis=1, retrieved July 2,
About the authors
Alexandra Branzan Albu is the IAPR Newsletter Editor since January 2009. She has attended all ICPR conferences since ICPR 2002. She has contributed papers to ICPR 2004, 2006, 2008, and 2010. She is looking forward to attending ICPR 2010.
David Kerr is a Master’s student with the ECE Department at the University of Victoria. He currently works under Dr. Branzan Albu’s supervision on Content-Based Image Annotation.
By Alexandra Branzan Albu (Canada) and David Kerr (Canada)