A priori minimization in pattern recognition. Application to industrial fish sorting and face recognition by computer vision

This paper deals with recognition of complex shape and the method used is based on minimization of prior knowledge. We study both representations of image features and object shapes linked to photometry and geometry properties. A fuzzy representation is used which is defined by independent membership functions of geometric distortion, such as rotation, zoom and deformations due to the complexity of natural objects. This method is applied in two problems: automatic fish sorting and face recognition, where objects are complex for being real-world images.