As we address the serious need of image classification tools for marine scientists, we note both specific and general challenges across these multiple imaging instruments. As we examine different scientific applications, the problems will vary as particle, cell, colony, or animal images as presented in different contexts. There will be a need to customize the features to be extracted from a wide pool of features. However, over the 25 years of a diverse set of research projects by the UMass Computer Vision Laboratory, we have experience with a very wide range of features for processing very different objects from a diverse set of domains. We will explore feature sets across color, texture, shape and size for use in these domains, and as they are found useful, they will be compiled into a library available for general use. Of equal importance to the quality and type of feature is the form of classifier. Over the five-years of effort, we will examine many different classifiers, and as we do they will be added to the library for available use in other applications. We believe the set of tools we plan to create here will have significant general utility, and will appeal to experts from various sub disciplines across marine science.

• Classification Methods

• Reference Image Database from a Living Phytoplankton Collection

• Multi-Scale Gaussian Differential Features for Appearance-Based Similarity Matching

• Use of Color Information as a Feature for Matching and Analysis

• Dual-Aspect Imaging in Flow

One of the first priorities in this project will be the development of expert classified image databases for each image type. This will be the primary task of the marine science expert sin the first two years. Development of classification algorithms for each image type will be the top priority for the computer scientists. As software tools become available they will be tested by the experts for the different imaging systems to provide iterative feedback to the software development effort. Classification development and testing will take place throughout the 5 years, since this forms the core of the project. In years 3 and 5 we will present our results at workshops held in conjunction with the national Ocean Sciences meeting. The first workshop will seek feedback for the development effort from the wider marine scientist community. The second workshop will be conducted as a tutorial to present our tools to potential users. The universe of possible combinations of imaging systems, marine science applications, particle types, classification tools, and image features we could consider is extremely large. We will apply our set of classification techniques to many different specific applications in marine science over the course of this project. However, we note a set of priority applications that will be our initial focus for each imaging system here.