Projects overview

Counting and classifying fish in their natural environment is a useful way of recovering important information. The advances of technology in the field of real time image processing make fish identification a very interesting area. This work combines image processing techniques such as color processing and feature extraction combined with case based reasoning and real time computing to classify species of fish which are similar to each other in both shape, size and color.

Identifying fish using image processing is not a trivial task. First of all, it must be ensured that the input is clear and error free. Then subject needs to be isolated that is, the horizon that separates seawater from seafloor must be identified so that backgrounds are removed. The fish isolation approach is radiacally different when the fish are in front of the seafloor in commparison to the case where they float in front of water. After this stage, the system attempts to mimic human recognition. For instance, human experts usually use a combination of features to distinguish among species. Experts take advantage of prior knowledge and experience. One way to replicate this expert behavior is to use case-based reasoning.

The architecture of our system includes a strucutred (tiered) architecture. The systems are capable to suppport real time opeation and the software relies on middleware that exposes functionality through differenciated presentation interfaces. The latter attempt to cover a wide range of use cases including smart devices, immersive interfaces and also streaming throught the web. Users may typically select a fish within a controlled environment and our system will take a snapshot of the selection, process it, identify the subject, recover relevant additional information and present it to the end-user according to the interface they are using.