Sorting salted cod fillets by computer vision

This pilot study reports an investigation into the potential for classification of salted cod fillets by analysis of colour images. The basis for classification was the Norwegian industry standard for salt fish fillets, which differentiates between classes imperial/superior and universal. The study was based on images of seven fillets of class imperial/superior and five of class universal selected by experienced workers. Different image analysis methods combined with SIMCA (soft independent modelling of class analogies) were investigated. Two of the methods, the 'red colour-variance' histogram and the grey-level histogram method, proved to be particularly promising. Each of these methods misclassified only one sample.