Information societies technology

C8. Economic development and scientific and technological prospects

Dissemination and exploitation plans are essential if the outcome of CATEGORIES are to be implemented and give birth to new economic activities. In this perspective, the Dissemination and Use Plan due at t+6 will be a major concern to all the partners, and all the more so as some ideas have already emerged:
Regarding the dissemination orientations, all the academic partners will be publishing their innovative research results in scientific conferences, scientific journals in the fields of computer vision, artificial intelligence, cognitive vision, geology and aerobiology. In addition, results will be disseminated through the dedicated web site and through the web sites of each partners, as well as through some national and European networks, among which the future European network of excellence in Cognitive Vision (ECVision, currently in preparation) and the European networks of aerobiology.

As for the exploitation of the results, the industrial partner (TIMEAT) has already expressed its interest in industrialising the outputs of the CATEGORIES project. Three main results are already expected to find an application in industrial activity:

• 
  A validated chitinozoans categorisation application
  
• 
  an integrated pollen grain categorisation system
  
• 
  a generic and reusable (hardware and software) modules for object categorisation.
  

A validated chitinozoans categorisation application
For a long time, the study of palaeontology has been forsaken by the prospection geology companies (minerals, petrol, etc), as they preferred using physical technologies. The characterisation of fossil appears to be a precious tool for datation and organisation of sediments. With these methods, the experts can give a datation with a precision of 1 Million year for 400 Million year old sediments. The experts in this domain became rare because their know-how requires a long and tedious experiment and manipulation time. The teaching of new researchers needs to aggregate an enormous database of pictures with a lot of rules of recognition. Categories could also simplify this teaching by unifying and harmonising one same database for all the research centres all around the world.
In addition, a second asset for microfossils recognition, is that there is an existing market of publics laboratories of micropaleontology and biology. This global market is about 40 systems on which our industrial partner plans to sale 15 systems to the more important teams.
Finally, this application will be also targeting the numerous petroleum companies worldwide by providing them with a reliable automated tools. In the past, some petroleum company had tried to make automatic identification system for chitinozoans identification, but they do not success because they do not dispose of advanced tools like the ones we will develop in this project. The European market of private laboratories of petroleum industries is equivalent to the one of public laboratories (15 systems). No evaluation of the market directly inside the platforms for exploration but it is strongly believed that it can be a precious tool for geologist which are far from the laboratories and with no expert at hand. In addition, if we succeed to get the scientific caution of this system, a global worldwide market will also be accessible. The CNRS-Geosciences’ involvement in CATEGORIES is a token of the quality of this application validation, as this partner is one of the major expert in this domain. It will moreover will participate to the dissemination activity of this application, hence supporting the final system with its own reputation and expertise.
The industrial adoption of a reliable automated chitinozoans categorisation system will be ensured by the fact that this technology will accuratelly provide the costly oil-drilling industry with essential information at reduced expenses.

Integrated Pollen Grain Categorisation System

In a first phase, it is planned to disseminate the pollen grain categorisation system built during the project among the palynologist community. There are about 300 sites in Europe that could be potentially interested in such a system allowing : -to reduce human labour necessary to read the slides

-to increase the number of sampling points.
This dissemination and exploitation plan will be made easier by the effective and active participation of the different laboratories involved in the end-user group. On top of this, the large European array of countries and regions (northern, eastern and Latin countries) among the end-user group is a key of success for the proposed solution to be accepted by the whole palynologist community.
This proposed system is going to offer a solution that will enable the users (palynological laboratories) in Europe to categorise a large variety of allergenic pollen types. The system will be focused from the beginning for Aerobiological studies applied to allergy, however, nowadays Aerobiological data can be also applied to the agronomic field. Airborne pollen should be a useful biological indicator to forecast the seed/fruit production in several particular varieties. On the other hand, the system can be adapted to be used in other fields of Palynology, such as Paleopalynology (studies on fossil pollen in soil) and Melissopalynology (studies on pollen for honey characterisation).
Regarding the recognition of pollen and as mentioned above, the global market is estimated to weight about 300 systems in Europe. TIMEAT’s objective in this field is to sale 100 systems in 3 years. This partner will be used as reference and gateway to commercialise this system. The promotion of this system in some major European country (France, Spain, Germany, Italy, UK) will be relayed through the project’s members . In addition it is believed that the global worldwide market is more that 500 systems on which it can be planned to access to 150 sales. On this last point, a partnership with a manufacturer of microscope station will have to be found in order to decrease sale costs in order to access different geographical markets
The European ASTHMA project has shown the feasibility and the usefulness to build a risk index for people suffering of asthma and other allergenic diseases. ACRI, the industrial partner of ASTHMA, has already developed a web server for disseminating in real-time forecast information on some allergenic pollen types and associated risk index. Currently, the in-situ measures of pollen concentrations are man made and this system strongly need a solution that can provide real-time and automatic pollen concentration measures. ACRI is strongly interested in adopting the integrated pollen grain categorisation solution proposed in the CATEGORIES project to enhance the existing server.

Generic and Reusable Modules for Object Categorisation

The generic solution developed in the CATEGORIES project is composed of three modules:

• 
  An image acquisition module for 3D microscopic objects
  
• 
  A knowledge acquisition and learning module for categorisation
  
• 
  An object categorisation module.
  

The economic potential of the CATEGORIES project lies with the technical and scientific originalities that will be developed. The control of an optical microscope to categorise 3D objects using cognitive vision represents a major economic advantage with respect to any other existing technique (confocal microscopy for example), usually more time consuming and less affordable. The scientific advantage of the system rests in the combination of skills in active vision, knowledge representation, learning techniques and 3D colour image processing.

The generecity that we intend to grant the system will make it usable in a wide range of applications. The concept of providing a user-friendly system allowing experts to define, in a convivial way, the classes of their application without any help from external computer scientists, sets this system ahead of any already existing systems. The robustness of the reasoning task in categorisation will make it suitable for various applications, first in the field of microscopy, but also in other fields where the acquisition device can be replaced without modifying either the knowledge or learning module, or the categorisation module.
In addition and in relation to vision activities, many tools to make control by vision in industries have been developed. For example, we are now able to measure, to detects defects, to control the position of different objects in mechanical, plastics et electronic industry. These tools permit to process the image, to make a classification using threshold on dimension, area, etc. However these controls are limited to the simple case on which the know-how of the operators is translated to a digital value (good or not good, dimension, etc). The field of investigation of vision for industry cannot answer today to problem concerning the classification of live objects or classification with the introduction of the know how of operators. We think that the technology of recognition developed in this project will open a new field of investigations for vision technology. Indeed, the developed system will permit to combine the data acquire by different sensors with know-how of experts. This do not exist today. Applications in food industries, biological laboratories will be solve : classification of raw material (seed, cells, …), aspect control on skin, tools for law management such as detection of bone trace in vegetables flour, etc. The Categories project is an major step of validation of the concept of automatic recognition. The access to the markets describe below will be done after a complete market study and search for funds and sales partners.
There is a great number of applications for which there are no operational system yet. These applications concern the control of minerals, living organisms, and all products for which classical geometry or rules do not apply. Other applications for petroleum industry and food industry will also have to be investigated for potential markets. Our categorisation approach is valid even if the objects studied are completely different.
The cognitive vision tools that will be developed in CATEGORIES will provide such solutions. TIMEAT has already expressed its interest in exploiting the different modules, either individually (one by one) or in an integrated solution.

C9 Annex C1 – References

Recognition and categorisation
[1] S. Ullman. High-Level Vision: Object Recognition and Visual Cognition. MIT Press, 1996.
[2] T. Poggio and S. Edelman. A network that learns to recognise three-dimensional objects. Nature, 343:263--266, 1990.
[3] A. J. Howell and H. Buxton. Learning identity with radial basis function networks. Neurocomputing, 20:15—34, 1998.
Knowledge representation
[4] E.A. Feigenbaum, A. Barr and P. Cohen (eds.)The Handbook of Artificial Intelligence. William Kaufmann, 1981.
[5] M. Minsky, A Framework for representing knowledge. In Psychology of Computer Vision, P.H. Winston (Ed.) McGraw-Hill, 1975.
[6] B. Draper, R. Collins, J. Brolio, A Hanson, E. Riseman The Schema System in Computer Vision: Advances and applications, R. Kasturi and R.J. Jain Eds, IEEE Press, 1991.
[7] M. Thonnat. The world of galaxies, chapter Toward an automatic classification of galaxies, pages 53--74. Springer-Verlag, 1989.
[8] M. Thonnat, V. Clement, and J.C. Ossola. Automatic galaxy description. Astrophysical Letters and Communication, 31(1-6):65--72, 1995
[9] S. Liu, M. Thonnat and M. Berthod. Automatic Classification of Planktonic Foraminifera by a Knowledge-based System. In IEEE CAIA, San Antonio, Texas, pp358-364, 1994.
[10] O. Capdevielle and P. Dalle. Image processing chain construction by interactive goal specification, IEEE ICIP, Austin ,USA, 1994.
[11] Blázquez, M.; Fernández, M.; García-Pinar, J.M.; Gomez-Perez, A. Building Ontologies at the Knowledge Level using the Ontology Design Environment KAW'98, Banff, Canada. 1998.
Learning
[12] M. I. Jordan. Learning in Graphical Models. NATO Science Series, vol.89, 1998.
[13] N. Christianini and J. Shawe-Taylor. An Introduction to Support Vector Machines. Cambridge University Press, 2000.
[14] P. Sollich. Bayesian methods for support vector machines: Evidence and predictive class probabilities, Machine Learning, 46:21-52, 2002.
[15] J. Pearl. Probabilistic Reasoning in Intelligent Systems : Networks of Plausible Inference. Morgan Kaufmann, 1988.
[16] K. P. Murphy. An introduction to graphical models, Technical Report, CS, UC Berkeley, 2001.
[17] A. K. Jain. Statistical pattern recognition: A review. IEEE Pattern Analysis and Machine Intelligence, 22:4—37, 2000.
[18] G. Hinton and Z. Ghahramani. Generative models for discovering sparse representations. Phil. Trans. Roy. Soc. B, 352:1177—1190, 1997.
[19] S. Roweis and Z. Ghahramani. A unifying review of linear Gaussian models. Neural Computation, 11:305—345, 2000.
[20] B.J. Frey and N. Jojic. Flexible models: A powerful alternative to exemplars and explicit models, in IEEE CVPR, Hawaii, (IEEE Press) 2001.
[21] C. Wallraven and H. Bultoff. Automatic acquisition of exemplar-based representations for recognition from image sequences, in CVPR, Hawaii, (IEEE Press) 2001.
[22] M. Mitchell. An Introduction to Genetic Algorithms. MIT Press, 1996.
[23] J. R. Koza. Genetic Programming: On the Programming of Computers by Natural Selection. MIT Press, 1992.
[24] J. R. Koza. Genetic Programming II: Automatic Discovery of Reusable Programs. MIT Press, 1994.
[25] S. Cognoni, A. B. Dobrzeniechi, R. Poli and J.C. Yanch. Genetic algorithm-based interactrive segmentation of 3D medical images. Image and Vision Computing, 17:881-895, 1999.
[26] W. B. Langdon and B. Buxton. Genetic Programming for combining classifiers, in GECCO, San Francisco, (Morgan Kaufmann) 2001.
[27] W. B. Langdon and B. Buxton. Combining decision trees and neural networks for drug discovery, in EuroGP, Ireland, (LNCS, Springer Verlag) 2002.
[28] Mallinson, H. and P.J. Bentley. Evolving fuzzy rules for pattern recognition, in International Conference on Computational Intelligence for Modeling, Control and Applications, 1999.
Task Control and Uncertainty
[29] M. Thonnat, V. Clement, and J. van den Elst. Supervision of perception tasks for autonomous systems: the OCAPI approach. Journal of Information Science and Technology, 3(2):140--163, 1994.
[30] T. Matsuyama. Expert systems for image processing: Knowledge based composition of image analysis processes Computer Vision Graphics Image Processes, 48:22--49, 1989.
[31] B. Draper, A. Hanson, and E. Riseman. Knowledge-directed vision: Control, learning, and integration. IEEE Signals and Symbols, 84(11):1625--1637, 1996.
[32] V. Clement and M.Thonnat. Integration of image processing procedures: OCAPI, a knowledge based approach. Computer Vision Graphics and Image Processing: Image Understanding, 57(2), 1993.
[33] S. A. Chien and H. B. Mortensen. Automating image processing for scientific data analysis of a large image database. IEEE Transactions on Pattern Analysis and Machine Intelligence, 18(8):854--859, 1996.
[34] R. Howarth and H. Buxton. Conceptual descriptions from monitoring and watching image sequences. Image and Vision Computing, 18:105—135, 2000.
[35] H. Buxton and S. Gong. Visual Surveillance in a dynamic and uncertain world. Artificial Intelligence, 78:431--459, 1995.
Vision techniques
[36] T. F. Cootes, G. J. Edwards, and C. J. Taylor. Active appearance models. IEEE Trans. Pattern Anal. Machine Intell., 23(6):681--685, 2001.
[37] J. M. Geusebroek, R. van den Boomgaard, A. W. M. Smeulders, and H. Geerts. Color invariance. IEEE Trans. Pattern Anal. Machine Intell., 23(12):1338--1350, 2001.
[38] J. M. Geusebroek, A. W. M. Smeulders, F. Cornelissen, and H. Geerts. Segmentation of tissue architecture by distance graph matching. Cytometry, 35(1):12--33, 1999.
[39] J. M. Geusebroek, A. W. M. Smeulders, and R. van den Boomgaard. Measurement of color invariants. In IEEE Conference on Computer Vision and Pattern Recognition, volume 1, pages 50--57. IEEE Computer Society, 2000.
[40] S. Ghebreab and A. W. M. Smeulders. Strings: variational deformable models of multivariate ordered features. Submitted to IEEE Trans. Pattern Anal. Machine Intell.
[41] B. Thai and G. Healey. Modeling and classifying symmetries using a multiscale opponent color representation. IEEE Trans. Pattern Anal. Machine Intell., 20(11):1224--1235, 1998.
[42] B. Thai and G. Healey. Optimal spatial filter selection for illumination-invariant color texture discrimination. IEEE Trans. Systems, Man, Cybernetics B, 30(4):610--615, 2000.
[43] J. van der Laak, M. M. Pahlplatz, A. G. Hanselaar, and P. C. de Wilde. Hue-saturation-density (HSD) model for stain recognition in digital images from Transmitted Light Microscopy. Cytometry, 39:275--284, 2000.
[44] J. O. Ramsay and R. W. Silverman. Functional Data Analysis. Springer-Verlag, Berlin, 1997.
[45] A.W.M. Smeulders, M. Worring, S. Santini, A. Gupta, and R. Jain. Content-based image retrieval at the end of the early years. IEEE Trans. Pattern Anal. Machine Intell., 22(12):1349--1380, 2000.
[46] H. M. G. Stokman and Th. Gevers. Error propagation due to color space transforms. In IST/SPIE Electronic Imaging, Machine Vision Applications in Industrial Inspection VIII, volume 3966. SPIE Press, 2000.

Pollen application

[47] K. D. Bennett. Pollen counting on a pocket computer. New Phytology, 114:275--280, 1990.

[48] W. R. Eisner and A. P. Sprague. Pollen counting on the microcomputer. Pollen et Spores, 29:461--470, 1987.
[49] O. Massot et al. The “C.Scope”: A new application to acquire and process simply and comfortably pollen count on PC. European Symposium on Aerobiology, Vienna, Austria, 2000.
[50] F. H. L. Benyon et al. Differentiation of allergenic fungal spores by image analysis, with application to aerobiological counts. Aerobiologia, 15:211—223, 1999.
[51] P. Li and J. R. Flenley. Pollen texture identification using neural networks. Grana, 38:59—64, 1999.
Chitinozoan application
[52] F. Paris, Y. Grahn, V. Nestor, I, Lakova, a revised chitinozoan classification. Journal of Paleontology 1999 PP549-570

Page of part C