A trainable n-tuple pattern classifier and its application for monitoring fish underwater

This paper describes a non-intrusive method for monitoring salmon stock. Biomass of individual salmon can be estimated remotely using salmon morphology and an underwater stereo imaging system. Salmon lateral length measurement could be measured by fitting a model to the fish in stereo images. However, the model fitting algorithm will need to be initiated manually by the user. Therefore an image processing technique that utilises a trainable n-tuple pattern recognition algorithm is under investigation. Provisional results of using the technique on a set of underwater salmon images are promising. Further experiment results show that the technique offers a fast and simple option for image segmentation and fish recognition in underwater images.