Using Sparse Representation for Fish Recognition and Verification in Real World Observation

The purpose of this paper is to present an innovated fish recognition and verification method suited for the real world automatic underwater fish observation. Based on the fish recognition and verification, biologists can study fish population as well as identify new species of fish appearing in area of interest. A distributed real-time underwater video stream system has been developed in Taiwan for long-term ecological observation. The system also archives video data and incorporates data analysis. We propose a fish detection procedure on the video data to obtain multiple species of fish images with varied angles, sizes, shapes, and illumination, which leads to a fish category database. In recent years, a sparse representation-based classification (SRC) based on compressive sensing is developed. Based on the SRC method, we propose a maximum probability of parting ranking method for fish recognition and verification, in which the eigenfaces and fisherfaces are used for feature extraction on the category database. Experimental results show that the proposed fish recognition and verification method is able to achieve high accuracy and robustness.