An associative neural network and its special purpose pipeline architecture in image analysis
There are several approaches to texture analysis and classification. Most have limitations in accurate discrimination or complexity in time calculation. A first phase is the extraction of texture features and later we classify it. Texture features should have the followings properties: be invariant under the transformations of translation, rotation, and scaling; a good discriminating power; and take the non-stationary nature of texture account. In Our approach we use Orthogonal Associative Neural Networks to Texture identification. It is used in the feature extraction and classification phase (where its energy function is minimized). Due his low computational cost and his regular computational structure the implementation of a real-time texture classifier based on this algorithm is feasible. There are several platforms to implement Artificial Neural Networks (VLSI chips, PC accelerator cards, multiboard computers, ...). The election relies on the type of neural model, their application, the response time, capacity of storage, type of communications, and so on. In this paper we present a pipeline architecture, where precision, cost and speed are optimally trade off. In addition we propose CPLD (Complex Programmable Logic Device) chips to complete realization of the system. CPLD chips have a reasonable density and performance at low cost.
TopicsComputer vision neural nets texture recognition real-time quality control
Unable to display preview. Download preview PDF.
- C. Neubauer, Segmentation of defects in textile fabric, Procc IEEE, pp. 688–691, 1992.Google Scholar
- Hoggar, S.G. Mathematics for Computer Science. Cambridge University Press. 1993.Google Scholar
- J.M. Zurada, Introduction to Artifial Neural Systems, West Publishing Company, 1992.Google Scholar
- Laws, K. Y. Texture Image Segmentation. Ph D. Thesis. University of Southern California. January. 1980.Google Scholar
- Francisco Ibarra Picó. Análisis de Texturas mediante Coeficiente Morfológico. Modelado Conexionista Aplicado. Ph. D. Thesis. Universidad de Alicante. 1995.Google Scholar
- Garcia-Chamizo J.M., Crespo-Llorente A. (1992) “Orthonormalized Associative Memories”. Proceeding of the IJCNN, Baltimore, vol 1, pg. 476–481.Google Scholar
- Hao J., Wanderwalle J. (1992) “A new model of neural associative memoriy” Proceedings of the JJCNN92, vol 2, pg. 166–171.Google Scholar
- Ibarra-Picó F., Garcia-Chamizo J.M. (1993) “Bidirectional Associative Orthonormalized Memories”. Actas AEPIA, vol 1, pg 20–30.Google Scholar
- Kosko, B. (1988a) “Bidirectional Associative Memories”. IEEE Tans. on Systems, Man & Cybernetics, vol 18.Google Scholar
- Kosko, B. (1988b) “Competitive adaptative bidirectional associative memories”. Procedings of the IEEE first International Conference on Neural Networks, eds M. Cardill and C. Butter vol 2. pp 759–766.Google Scholar
- Pao You-Han. (1989) “Adaptative Pattern Recognition and Neural Networks”. Addison-Wesley Publishing Company, Inc. pg 144–148.Google Scholar
- Wang, Cruz F.J., Mulligan. (1990b) “Two Coding Strategies for Bidirectional Associative Memory”, IEEE Tans. on Neural Networks, pg 81–92.ang, Cruz F.J.Google Scholar