Con esta herramienta te facilitamos un acceso a todas las ofertas y demandas de tecnología europeas y a búsquedas de socios para participar en propuestas europeas de I+D publicadas en la red Enterprise Europe Network, pudiendo filtrar los resultados para facilitar las búsquedas más acordes con tus necesidades.

¿Quieres recibir estos listados de oportunidades de colaboración en tu correo de forma periódica y personalizada? Date de alta en nuestro Boletín

Para optimizar los resultados de la búsqueda, se recomienda utilizar términos en inglés.

Desarrollo de una tecnología para mejorar la clasificación de objetos destinada a sistemas de reconocimiento de patrones

Resumen

Tipo:
Oferta Tecnológica
Referencia:
TORU20140407001
Publicado:
18/06/2015
Caducidad:
24/12/2015
Resumen:
Una empresa rusa especializada en I+D en el campo de análisis de imágenes y vídeos ha desarrollado una tecnología eficiente para mejorar la clasificación de objetos (uplearning). Los clasificadores actuales permiten reconocer miles de clases distintas de objetos gráficos en imágenes digitales pero la actualización del clasificador para reconocer una o más clases adicionales conlleva mucho tiempo. Esta tecnología permite reducir tiempos y mejorar la escalabilidad de sistemas de reconocimiento de patrones. Se buscan socios interesados en continuar con el desarrollo y establecer acuerdos de cooperación técnica y servicio.

Details

Tittle:
Development of a classifier uplearning technology for pattern recognition systems
Summary:
A company from the Moscow area (Russia), specialized in R&D in image and video analysis, developed an efficient technology for object classifier retraining (uplearning). This technology allows reducing the classifier training time and considerably increasing the scalability of pattern recognition systems. The company is looking for joint further development in various forms (technical cooperation, service agreement).
Description:
Increasing the number of classes, which a classifier algorithm is able to recognize, is an actual problem. Modern classifiers are able to recognize thousands different classes of graphics objects in digital images (man, car, dog, lion etc.). But updating of the classifier to make it recognize one or more additional classes takes a lot of time.
Presently, this problem is solved in the following steps: 1) classifier parameters are changed; 2) a new image sample set is created that includes all images of all classes (old and new); 3) a long process of new classifier training is performed.
The offered new approach for classifier retraining using images of only new objects and automatic tuning of the uplearning process parameters allow a considerable reduction of the classifier retraining time.
This approach makes it possible to design expandable and well scalable pattern recognition systems. Such systems are able to adapt quickly to variations of operating conditions.
The offered technology is suitable for different classifier types: artificial neural networks, support vector machine classifier and decision trees.
The company is looking for partners specialized in information technologies for joint further development; the partner sought can be an R&D organisation, a small/medium-sized company or individual developer.
Advantages and Innovations:
The novelty of the developed technology is in the classifier uplearning algorithm that takes into account the classifier's internal state (its parameters), and also the quality of the training samples.
The advantages of the technology are:
· 50% reduction of the time for creation of a new classifier based on old images samples and old classifier;
· improving the adaptation ability of pattern recognition system, due to the possibility of adding new classes of objects without complete classifier retraining. As result, about 90% reduction of the time needed for introducing new classes into the classifier
· independence of the technology from the kind of the pattern (audio, visual, text and other) to be recognized.
Stage of Development:
Under development/lab tested
IPs:
Secret know-how
CommeR Statunts Regarding IPR Status:
RF patent application is planned to be filed.

Partner sought

Type and Role of Partner Sought:
The partner type: small/medium-sized company, R&D organisation or individual developer qualified in mathematical programming, object oriented programming, experienced in solving application task in pattern recognition area. Having a high-performance computing architecture (network cluster, mainframe) will be a big advantage.
The partner's area: information technologies, graphics software.
The partner's role: provision of production resources required for development, testing and further improvement of the technology; funding of the joint development.

Client

Type and Size of Client:
Industry SME <= 10
Already Engaged in Trans-National Cooperation:
No
Languages Spoken:
English
Russian

Keywords

Technology Keywords:
01005005 Filtrado de información, semántica, estadística
01003012 Imagen, procesado de imágenes, reconocimiento de modelos
01006012 Informática aplicada a descripción de imágenes y vídeo
01003003 Inteligencia artificial (IA)
01003007 Tecnología informática / gráficos, meta informática