This report describes the research conducted during the master thesis for the Vision, Graphics and Interactive Systems master programme. In this work, we explore a part of animal tracking: re-identification of zebrafish (Danio rerio), in particular. The literature demonstrates the importance of the zebrafish for drug development and pushing the human understanding of organisms further. The state-of-the-art re-identification methods of zebrafish employ solely a top-view and grayscale images and require a lot of data for the best performance. This research rather aims to study how well the side-view and color images can perform in the same task while keeping data processing to a minimum. Inspired by the person re-identification field, two feature descriptors, each encoding both color and texture information, and five metric learning algorithms were tested. The contribution of the color and texture components of the feature descriptors were also investigated. The experiments were conducted on two and three fish, separately. The results show that for the same fish, a mean average precision (mAP) of 100% can be obtained, and when testing on previously unseen fish, a mAP of 89.40% is achieved, at best. When training on small sequences, with a length up to only 50 frames, the methods can sustain very few (0-2%) identity switches. This approach clearly shows potential and encourages the further research on the topic.