PDL-HM - MORPHOLOGICAL AND SYNTACTIC SHAPE CLASSIFICATION ALGORITHM - REAL-TIME APPLICATION TO FISH SPECIES CLASSIFICATION

This paper describes a novel shape classification technique in computer vision called PDL-HM, which combines the selection of morphological structuring elements from contour description languages, the extraction of shape characteristics by the hit-and-miss transform, and shape classification by a neural network. This real-time method is orientation independent and tolerates limited object overlap. It is illustrated by an application to fish species classification in an industrial environment and performs excellently.