| 摘要 |
Although an image contains a large amount of information, it is usual to scan the entire image in the form of a raster, for example by means of a television camera, and to enter it into a computer for image analysis or pattern recognition. The visual system of living beings selects the information required in each case by means of saccadic eye movements and thus considerably reduces the flow of information to the brain. Similarly, an object contour is found according to the invention, traced and segmented at the point where contour parameters, for example the curvature, change. Data on contour elements describe nodes on a graph, their positional relation describes the edge between two nodes. By comparison with previously learned graph structures, hypotheses are formed on the type or class of the pattern, and on the position of other contour elements and/or segmenting points of the object hypothesis. The computer drives the sensor to these locations in a saccadic movement and checks the hypothesis in order to either reinforce it or to weaken it. After a few saccadic movements, one hypothesis dominates. The invention is extended to other measurement data and evaluating methods.
|
