The joint IAPR International Workshops on Structural and Syntactic Pattern Recognition (SSPR 2012) and Statistical Techniques in Pattern Recognition (SPR 2012) were held at Miyajima-Itsukushima, Hiroshima, between 7th-9th November, 2012. These are respectively the 14th and 9th editions of  the SSPR and SPR workshops. This joint event is biannually organized by Technical Committee 1 (Statistical Pattern Recognition Technique) and Technical Committee 2 (Structural and Syntactical Pattern Recognition) of the  International Association of Pattern Recognition (IAPR), and held in conjunction with the International Conference on Pattern Recognition (ICPR).

In front of the conference hotel in the suburbs of the city of Hiroshima, we could see the Grand Gate (Otori) of Itsukushima shrine, which is listed as a UNESCO World Heritage Site.

As is now tradition, during the SPR workshop, the Pierre Devijver Award recipient presents an invited lecture. The 2012 award winner was Professor George Nagy from Rensselaer Polytechnic Institute in Troy, New York, USA. The workshop also contained invited talks by Kenichi Kanatani from Okayama University and Ales Leonardis from the University of Birmingham.

The topics of the SSPR Workshop were Structural Matching and Syntactic Method; Probabilistic and Stochastic Structural Models; Graphical Models and Graph-Based Models; Spectral Methods for Graph-Based Representations; Kernal Methods for Structured Data; Structural Learning in Spatial or Spatio-Temporal Signals; SSPR Methods in Text, Document, Shape Image, Video and Multimedia Signal Analysis; Intelligent Sensing Systems; and Novel Applications.

The SPR Topics were Multiple Classifiers and Large Margin Classifiers; Density Estimation and Model Selection; Ensemble Methods, Bayesian Methods and Kernal Methods; Independent Component Analysis and Compressed Representation; Unsupervised and Semi-Supervised Learning; Linear and Non-linear Manifold Learning; Gaussian Processes; Dimensionality Reduction; Cluster Analysis; Data Visualization; Hybrid Methods; Comparative Studies; SPR Methods in Text, Document, Shape Image, Video and Multimedia Signal Analysis; and Novel Applications.

There were some 120 papers submitted to the joint workshops and 80 papers from 18 countries were accepted. We thank the members of the international program committee for their thoughtful reviews, which have led to the interesting and varied set of papers presented at the conference and published in the proceedings.

The main social event of the workshop was a visit to Itsukushima. At the island, participants enjoyed the twilight on the red shrine from the shore and lighted from west in the afternoon, like living in a movie with the Emperor (Twilight in the Forbidden City by Sir Reginald Fleming Johnston, The Last Emperor by Bernardo Bertolucci). They also learned the Japanese middle age theology as a combination Shinto and Buddhism.

In 2012, the joint SSPR and SPR Workshops were co-hosted by the pattern recognition research groups from four Japanese Universities, namely Hokkaido University, Tohoku University, Hiroshima University, and Chiba University. The Special Interest Group of Pattern Recognition and Media Understanding (SIG PRMU, formerly SIG PR) of the Institute of Electronic Information and Communication Engineers (IEICE) of Japan also gave formal support for this event. SIG PRMU(PR) is one of the oldest communities for pattern recognition in the world, dating back to 1960s. Interestingly, the origins of the Principal Component Analysis (PCA) technique now universally used in pattern recognition can be traced back to independent early work by Taizo Iijima (1963, at the former Electrotechnical Laboratory of MITI) and Satosi Watanabe (1962, from the University of Hawaii). Today PCA is an indispensable tool in pattern recognition that has recently been extended to give both sparse and kernel methods, providing powerful new tools for data reduction. In the 1970’s basic methodology from structural and syntactical pattern recognition were used in a national project concerned with “Kanji” (Chinese Character used in Japanese context) character recognition and the results presented and discussed at a historically significant meeting of SIG PR.

We gratefully acknowledge financial support from the Institute of Media and Information Technology, Chiba University and from Chiba University. We also acknowledge valuable support from Hokkaido University, Tohoku University, Hiroshima University, and the Special Interest Group of Pattern Recognition and Media Understanding in Institute of Electronic Information and Communication Engineers of Japan.

The next S+SSPR will be organized by Professor Pasi Fränti of the University of Eastern Finland, Jouensuu, in the city of Joensuu, Finland, on August 20-24, 2014, one week before the 22nd ICPR at Stockholm.

Workshop Report:  S+SSPR 2012

Report prepared by Atsushi Imiya (Japan)

Co-Organizers:

 

Atsushi Imiya, IAPR Fellow (Japan)

Mineichi Kudo, IAPR Fellow (Japan)

Georgy Gimel’farb (New Zealand)

Keiji Yamada (Japan)

Proceedings of the

workshop have been

published by Springer

in the series

Lecture Notes in

Computer Science

(Volume 7626)

Co-Program Chairs:

 

Arjan Kuijper (Germany)

Edwin Hancock, IAPR Fellow (UK)

Shinichiro Omachi (Japan)

Terry Windeatt (UK)

George Nagy, Pierre Devijver Award Lecture

“Estimation, Learning, and Adaptation:  Systems that Improve with Use”

Abstract:

The accuracy of automated classification (labeling) of single patterns, especially printed, hand-printed, or handwritten characters, leveled off some time ago. Further gains in accuracy depend on classifying unordered sets or ordered sequences of patterns. Linguistic context, already widely used, relies on 1-D lexical and syntactic constraints in plain text. Style-constrained classification exploits the shape-similarity of sets of same-source (isogenous) characters of either the same or different classes. 2-D structural and relational constraints are necessary for understanding tables and forms. Applications of pattern recognition that do not exceed the limits of human sensory and cognitive systems can benefit from green interaction whereby operator corrections are incorporated into the classifier.

Kenichi Kanatani

“Optimization Techniques for Geometric Estimation:  Beyond Minimization”

Abstract:

We overview techniques for optimal geometric estimation from noisy observations for computer vision applications. We first describe estimation techniques based on minimization of given cost functions: least squares, maximum likelihood, which includes reprojection error minimization as a special case, and Sampson error minimization. We then formulate estimation techniques not based on minimization of any cost function: iterative reweight, renormalization, and hyper-renormalization. Showing numerical examples, we conclude that hyper-renormalization is robust to noise and currently is the best method.

Ales Leonardis

“Hierarchical Compositional Representations of Object Structure”

Abstract:

Visual categorisation has been an area of intensive research in the vision community for several decades. Ultimately, the goal is to efficiently detect and recognize an increasing number of object classes. The problem entangles three highly interconnected issues: the internal object representation, which should compactly capture the visual variability of objects and generalize well over each class; a means for learning the representation from a set of input images with as little supervision as possible; and an effective inference algorithm that robustly matches the object representation against the image and scales favorably with the number of objects. In this talk I will present our approach which combines a learned compositional hierarchy, representing (2D) shapes of multiple object classes, and a coarse-to-fine matching scheme that exploits a taxonomy of objects to perform efficient object detection. Our framework for learning a hierarchical compositional shape vocabulary for representing multiple object classes takes simple contour fragments and learns their frequent spatial configurations. These are recursively combined into increasingly more complex and class-specific shape compositions, each exerting a high degree of shape variability. At the top-level of the vocabulary, the compositions represent the whole shapes of the objects. The vocabulary is learned layer after layer, by gradually increasing the size of the window of analysis and reducing the spatial resolution at which the shape configurations are learned. The lower layers are learned jointly on images of all classes, whereas the higher layers of the vocabulary are learned incrementally, by presenting the algorithm with one object class after another. However, in order for recognition systems to scale to a larger number of object categories, and achieve running times logarithmic in the number of classes, building visual class taxonomies becomes necessary. We propose an approach for speeding up recognition times of multi-class part-based object representations. The main idea is to construct a taxonomy of constellation models cascaded from coarse-to-fine resolution and use it in recognition with an efficient search strategy. The structure and the depth of the taxonomy is built automatically in a way that minimizes the number of expected computations during recognition by optimizing the cost-to-power ratio. The combination of the learned taxonomy with the compositional hierarchy of object shape achieves efficiency both with respect to the representation of the structure of objects and in terms of the number of modeled object classes. The experimental results show that the learned multi-class object representation achieves a detection performance comparable to the current state-of-the-art flat approaches with both faster inference and shorter training times.

Click here to go to the publisher’s

web site for the volume.

Photo by Arjan Kuijper

Click here to return to Contents Page