Efficient Instance-based Fish Species Visual Identification by Global Representation

This paper presents the participation of the LSIS/DYNI team for the ImageCLEF 2014 Fish identification challenge. ImageCLEF's Fish identification task provides a testbed for the system-oriented evaluation of fish species identification based on still images. The goal is to investigate image retrieval approaches in the context of images extracted from collected videos. The LSIS/DYNI team submitted three runs,, won the challenge with re- sults that sensibly outperform the baseline (both recall and precision of 0.99) for the image- based fish recognition category with a fully automatic method. Our approach is based on a computer vision framework involving local, highly discriminative visual descriptors, so- phisticated visual-patches encoder and large-scale supervised classification. The paper pre- sents the three procedures employed, and provides an analysis of the obtained evaluation re- sults.