TH E CN N I MPLEMEN TATI ON OF WAVE TYPE METRI C FOR I MAGE AN ALYSI SAN D CLASSI FI CATI ON
István Szatmári, Tamás Roska · 1998
I n this paper a CN N based wave type metric is discussed and designed for object classification. The autowave metric, a nonlinear variant of the H ausdorff metric, is used. This approach turned out to be superior compared to some other classification methods, e.g. the H amming distance calculation. A number of tests have been completed within the so- called bubble/ debris segmentation experiments using original and artificial gray-scale images (13, 14). H ere, we show the details of the CN N implementation and discuss its properties. The single-layer trigger wave generation and the two-layer implementation of wave type metric results in a flexible and efficient tool for object classification. The VLSI complexity of the proposed solution is also analyzed.