The work presented in this thesis proposes computational models for motion extraction and pattern recognition from the visual flow of information in the brain and determine how the two tasks can be understood together. More precisely, we propose hypotheses about how the brain mechanism for these tasks may work and we seek to show how neurons with a small receptive field are able to deliver coherent answers and encode complex patterns. We study each aspect of brain processing that we have modelled in a connectionist framework, showing how such distributed systems can be used in complex tasks such as motion detection and pattern recognition. From the computer science perspective, these models provide new algorithms, with interesting properties such as distributed memory utilization and robustness. We have focused our work on two aspects of visual information processing: the detection of motion and the discrimination of complex visual patterns from that signal (or "cognitive vision") that constitute the two parts in which this work is divided. The first part of this thesis about motion detection studies how feature extraction is performed from the visual flow of information, specifically how problems due to the small size and range of the local motion detectors can be solved. In the second part, we work on how the classification of complex visual patterns is achieved from the processing provided by the early vision system, evaluating different feature-extraction techniques to perform what we call cognitive vision.

EXPLOR_STEP=$WICRI_ROOT/Wicri/Lorraine/explor/InforLorV4/Data/Main/Exploration

HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 002058 | SxmlIndent | more

HfdSelect -h $EXPLOR_AREA/Data/Main/Exploration/biblio.hfd -nk 002058 | SxmlIndent | more

{{Explor lien
   |wiki=    Wicri/Lorraine
   |area=    InforLorV4
   |flux=    Main
   |étape=   Exploration
   |type=    RBID
   |clé=     Hal:tel-00642818
   |texte=   Distributed neural computation for the visual perception of motion
}}