Image-Based Recognition of Individual Trouts in the Wild

Abstract

Individual fish recognition has potentials in applications as fish cultivcation and fishing tourism. Unlike previous research, which either based on physical marker or based on photograph comparison using observers, this paper propose an approach being able to identify individual brown trouts (Salmo trutta) automatically with images taken in the wild. Although big variation in illumination, poses of the trouts, and resolution we validated that just using a small patch taken from the head of the trout, which can minimize the variations, it's possible to recognize individuals automatically. Two methods were proposed based on a local density profile and on a codebook. Both of the methods gave modest recognition accuracy 64.9% and 74% respectively, which compared to random chance at 3.3% is significantly better.