A Machine Learning Tool for Image Quality Assessment through Prediction of Iris Recognition Success

In Phase I we developed a machine learning method for predicting match score errors in iris imagery to determine the quality of the image. We now propose to develop a flexible and configurable tool for creating, refining, and applying such a match score predictor to any image quality assessment problem where a training signal can be identified. The tool will enable users to supply ground truth for image labeling through an intuitive plugin. It will provide a set of feature plugins for feature extraction while allowing users to add their own. It will provide users with access to the training process for the image quality assessment through a training plugin. It will offer automatic feature selection to reduce the feature set to the most efficacious through a feature selection plugin. And it will provide extensive analysis capabilities to determine the effectiveness of the image quality assessor on test data. Scripting support will allow the user to invoke our algorithms without need for the GUI if desired. Such a flexible image quality assessment system will have application beyond iris recognition to other areas in biometrics, such as face recognition, but also to domains such as stereovision, visual odometry, and general object recognition. Our work will take this technology to TRL 4 toward TRL 5 through integration of commercial segmentation software, testing on realistic data, and interfacing with an iris imager to simulate the target environment.