Fish fillet authentication by image analysis

Abstract

The work aims at developing an image analysis procedure able to distinguish high value fillets of Atlantic cod (Gadus morhua) from those of haddock (Melanogrammus aeglefinus). The images of fresh G. morhua (n = 90) and M. aeglefinus (n = 91) fillets were collected by a flatbed scanner and processed at different levels. Both untreated and edge-based segmented (Canny algorithm) regions of interest were submitted to surface texture evaluation by Grey Level Co-occurrence Matrix analysis. Twelve surface texture variables selected by Principal Component Analysis or by SELECT algorithm were then used to develop Linear Discriminant Analysis models. An average correct classification rate ranging from 86.05 to 92.31% was obtained in prediction, irrespective the use of raw or segmented images. These findings pave the way for a simple machine vision system to be implemented along fish market chain, in order to provide stakeholders with a simple, rapid and cost-effective system useful in fighting commercial frauds.