Predicting Performance of a Face Recognition System Based on Image Quality
This work addresses the need for preemptive performance assessment in face recognition systems, though it is incremental as it builds on existing generative modeling with a Bayesian approach to handle limited training data.
The authors tackled the problem of predicting face recognition system performance using image quality features, developing a generative model that accurately forecasts performance before recognition occurs, with experiments on six systems across three datasets showing accurate predictions when using an unbiased Image Quality Assessor.
In this dissertation, we present a generative model to capture the relation between facial image quality features (like pose, illumination direction, etc) and face recognition performance. Such a model can be used to predict the performance of a face recognition system. Since the model is based solely on image quality features, performance predictions can be done even before the actual recognition has taken place thereby facilitating many preemptive action. A practical limitation of such a data driven generative model is the limited nature of training data set. To address this limitation, we have developed a Bayesian approach to model the distribution of recognition performance measure based on the number of match and non-match scores in small regions of the image quality space. Random samples drawn from these models provide the initial data essential for training the generative model. Experiment results based on six face recognition systems operating on three independent data sets show that the proposed performance prediction model can accurately predict face recognition performance using an accurate and unbiased Image Quality Assessor (IQA). Furthermore, our results show that variability in the unaccounted quality space -- the image quality features not considered by the IQA -- is the major factor causing inaccuracies in predicted performance.