Neural network arbitration for edge detection

Our proposal is that different edge detection techniques be utilised in parallel, for generalised edge detection of different image types. This allows the attainment of several edge maps containing different features. From the various edge maps, edges are arbitrated in order to perform the merging of the different maps. The arbitration is done using a perceptron neural network. To overcome the increased computing time, implementation is being performed on a multi-transputer array, using the inherent parallelism of the techniques involved. In this paper, we present examples that show the performance of the neural arbitration of different techniques along with a comparison with the individual techniques alone. These are computed and calculated using test patterns which allow for easy quantification